下载文档原格式
英语情景对话：英文问路指路A: Excuse me, Where am I on this map?B: We are here, bus station, we are in the heart of the city.A: Oh ! I think I’m lost. Can I go from here to the railway station?B: Head straight up the street about two blocks then turn left.A：对不起，请问我在地图上的什么地方?B：我们在这里，汽车站，我们现在在市中心。
A：哦！我想我迷路了。
我能否从这里到火车站呢？B：顺这条街一直走过两个街区，然后左转。
A: Excuse me. I’m afraid I got lost. Can you show me the way to the station?B: I’m walking that way. Let me lead you the way.A：对不起，我迷路了，请问您能告诉我去车站怎么走吗？B：我正朝那边去。
让我给你带路吧！A: Exc use me. I wonder if you could help me. I’m looking for the Museum.B: Boy, you are lost. It’s across town.A: Oh ! What bad luck ! How can I get to the Museum?B: You can take a No. 24 bus here and then transfer to a No.53 bus to get there.A：对不起，打扰一下，不知您能否帮助我，我在找博物馆。
B：哇，你是迷路了。
它在城的那头。
A：哦！太糟糕了！那我怎么去博物馆呢？B：您可以在此乘坐24路公共汽车换乘83路公共汽车到那里。
保持交通安全的英语作文80词,七年级全文共6篇示例，供读者参考篇1Keeping Safe on the Roads and StreetsHey everyone! I'm here to talk to you about something super important - traffic safety. As kids, we're out there on the roads and streets all the time, walking to school, riding our bikes, or just playing around the neighborhood. But it's crucial that we stay safe and avoid accidents.First up, let's cover pedestrian safety. Whenever you're walking near roads, you've got to pay close attention. Don't get distracted by your phone or music. Look both ways before crossing, and keep an eye out for cars. Use crosswalks and obey traffic signals when they're available. If there's no crosswalk, find a spot with good visibility and walk across when there's a big enough gap in traffic.Drivers, please be patient and watch out for pedestrians, especially kids. We can be unpredictable and might not know all the rules yet. Slow down in residential areas and near schools. Yield to people crossing the street.For those of us who ride bikes, we've got some extra safety steps. Helmet on, every ride! A helmet could literally save your life if you fall or get hit. Obey all traffic signs and signals, just like you would driving a car. Use hand signals when turning. Ride predictably in a straight line, don't swerve between cars.Ride on the street in the same direction as traffic, not against it. Use bike lanes when available. If there's no bike lane, ride as far to the right as possible. Be extra careful at intersections - that's where a lot of accidents happen with bikes.And drivers, share the road! Give cyclists at least three feet of space when passing. Check your blind spots before turning or changing lanes. We're harder to see than a car, so stay alert.Now for some general street smarts for kids: Only cross at corners, not mid-block. Make eye contact with drivers before you cross, even if you have the right of way. Watch for cars backing up or turning. Bright clothes and reflective gear help drivers see you better, especially at night.If you get on or off a bus, use the door on the curb side, not the traffic side. Wait for the bus to come to a complete stop before approaching it.This one's for my fellow scooter and skateboard riders - wear that helmet and safety gear! Skate on smooth, paved areas without traffic, like skate parks. Don't skate in the street!For kids who like to play sports in the street, don't ever chase balls into the road without an adult checking first. It's just not worth the risk.And a reminder for drivers - follow the speed limits, especially in neighborhoods and school zones. Put that phone down and avoid distractions. Be patient around kids - we're still learning.At the end of the day, traffic safety is everyone's responsibility. We all want to get where we're going safely. So pay attention, follow the rules, and look out for each other.For kids, listening to your parents and teachers about street smarts is so important. We're little, they're bigger - it's harder for drivers to see us. The roads can be really dangerous if we're not being safe and making good choices.I hope these tips help you stay safe and sound, whether you're walking, biking, skating, or just playing outside. Be street smart, be alert, and make good decisions. Let's all do our part to stay safe and get where we need to go in one piece!篇2Keeping Safe on the Roads and StreetsHi there! My name is Emma and I'm a 7th grader. Today I want to talk to you about something really important - traffic safety. Whether you're walking, riding your bike, or just playing outside, it's super duper important to be safe around cars, trucks, and other vehicles. Let me tell you why!First up, let's talk about walking safely. Whenever you're walking near the street, you've got to use crosswalks and obey the traffic signals. That means waiting for the "Walk" sign and looking both ways before crossing. My mom is always reminding me to make eye contact with drivers before stepping out, just to be extra sure they see me. If there's no crosswalk or signal, you should walk facing traffic so you can see any cars coming.And don't forget, it's safest to walk on sidewalks or paths whenever possible. If there are no sidewalks, you need to walk as far away from the road as you can. My little brother Jacob learned that one the hard way when he was walking on the street and almost got clipped by a passing car! Luckily my dad was with him and pulled him back just in time. Phew!Now let's chat about riding your bike, scooter, or skateboard. The basic rules are pretty similar to walking - you've got to obey all traffic signs and signals, and ride cautiously near vehicles. But there are some extra safety tips just for riders.First of all, you must, must, MUST wear a helmet every篇3Traffic Safety is Super Important!Hi everyone! My name is Jamie and I'm in 7th grade. Today I want to talk about something that's really important - traffic safety! Whether you're walking, riding your bike, or getting a ride in the car, we all need to be careful and follow the rules to stay safe on the roads.Let's start with walking. I know it can be tempting to cross the street wherever you want, especially if you're in a hurry. But that's really dangerous! You could get hit by a car and seriously hurt. Always use the crosswalks and obey the traffic signals. Look both ways before crossing, even if you have the walk signal. And avoid distractions like your phone when crossing. A driver might not see you if you're not paying attention.Riding your bike is fun, but you have to be responsible. Always wear a helmet - even if it's not cool looking, it could save your life! Obey all traffic lights and signs just like you would in a car. Use hand signals when turning so drivers know what you're doing. Never ride against traffic or weave in and out between parked cars. Be predictable and follow the rules of the road.If you're getting a ride in the car, buckle up! Your seatbelt is your best defense in a crash. Even if it's just a short trip, wear your seatbelt every time. Don't distract the driver by being too loud or moving around a lot in the backseat. If the driver needs to focus, give them some quiet time. And of course, never stick any body parts out the window!Driving is a huge responsibility that takes practice. That's why there are laws about when you can get a permit and license. Pay close attention in driver's ed so you learn the rules. Texting, eating, putting on makeup - anything that takes your eyes off the road is distracted driving and can be deadly. Drunk driving is incredibly dangerous too, and you should never ever do it.Following traffic laws isn't just about avoiding tickets. It's about keeping yourself and others safe. Pedestrians, cyclists, and drivers all have to share the roads and look out for each other. Even if you're just walking down the street, you could encountera car at any moment. Being predictable and following traffic signals helps drivers anticipate what you'll do next.I know it's easy to get impatient, especially if you're in a hurry. But rushing and breaking rules is how accidents happen. A few seconds of your time is way better than putting your life at risk! Pay attention, avoid distractions, and respect the traffic laws.Cars are big heavy machines that can cause a lot of damage if they're not operated properly. Just think about how much force is involved when a couple ton vehicle crashes into something! That's why we have to be so careful with traffic safety from a very young age.My篇4Staying Safe on the Roads: A Kid's Guide to Traffic SafetyHey there, kids! Safety Sam here to talk about something super important – keeping safe when you're out and about near roads and traffic. I know it might not seem like the most exciting topic, but trust me, it's way better than ending up as a road pancake!First things first, let's talk about crossing the street. I know, I know, it seems pretty straightforward, but you'd be surprised how many kids get into sticky situations because they're not being careful. Here are some golden rules for crossing safely:Always use a crosswalk or intersection if there is one. Those zebra stripes aren't just there for decoration!Look left, right, and left again before you step into the road. You never know when a car might come zipping around the corner.If there are traffic lights, wait for the "Walk" signal before crossing. Don't be a daredevil and try to beat the light – that's just asking for trouble.Keep your eyes and ears open for any approaching vehicles, even if you have the right of way. Some drivers can be real space cadets behind the wheel.If you're with a group, hold hands and stay together. Safety in numbers, amigos!Next up, let's talk about being a pedestrian in general. Even if you're not crossing the street, there are still plenty of ways to get yourself into a pickle if you're not paying attention. Here are some key tips:Use sidewalks whenever possible. If there aren't any sidewalks, walk facing traffic so you can see any approaching vehicles.Stay alert and keep your eyes on the road, not buried in your phone or other gadgets. You don't want to become a real-life version of those zombie games you love so much!Wear bright or reflective clothing, especially at night or in low-light conditions. You want drivers to be able to see you from a mile away.Avoid walking near the edges of roads or highways if possible. Those are prime real estate for getting clipped by a car or truck.If you're walking in a group, walk single file and stay close to the edge of the road or sidewalk.Now, let's talk about being a passenger in a vehicle. Just because you're not the one driving doesn't mean you can totally zone out. Here are some tips for staying safe while riding along:Always wear your seatbelt, no exceptions. It's not just the law – it could literally save your life in a crash.Avoid distracting the driver with loud music, rowdy behavior, or constant chatter. Let them focus on the road.If you're in the front seat, adjust your seat and headrest properly for maximum safety.Keep your arms, legs, and other body parts inside the vehicle at all times. You don't want to lose a limb to a passing car or tree branch!If the driver seems impaired or is driving recklessly, speak up or ask them to pull over. Your safety is more important than hurting their feelings.Last but not least, let's cover some general safety tips that apply whether you're walking, biking, or just hanging out near roads:Obey all traffic signals, signs, and crossing guards. They're there for a reason, folks.Stay alert and avoid distractions like headphones or your phone when near traffic.If you're riding a bike, scooter, or skateboard, wear a helmet and follow the same rules as pedestrians and vehicles.Avoid playing or hanging out in or near the street, driveways, or parking lots. Those are prime accident zones.If you ever feel unsafe or uncomfortable in a traffic situation, get to a safe place and tell a trusted adult right away.There you have it, kids – the ultimate guide to staying safe and sound around traffic. Follow these tips, and you'll be a regular road warrior in no time! Remember, it's always better to be safe than篇5Keeping Our Roads Safe: A Kid's Guide to Traffic SafetyHi there! I'm just a regular 7th grader, but I've got some important stuff to share about traffic safety. You might be thinking, "Ugh, another boring lecture about looking both ways before crossing the street." But trust me, this is way more exciting than that!First up, let's talk about why traffic safety matters. Every year, thousands of people get seriously injured or even killed in car accidents. That's just not cool. We're all part of the same community, and we need to look out for each other on the roads. By being safe and responsible, we can help prevent those terrible tragedies.Now, I know what you're thinking: "But I'm just a kid! What can I do?" Well, buckle up (get it?), because I've got some awesome tips for you, whether you're a pedestrian, a cyclist, or a passenger in a car.For my fellow pedestrians out there, listen up! Always use crosswalks and obey traffic signals. Don't just randomly dart across the street like a superhero – you're not invincible, my friend. And when you're waiting to cross, step back from the curb and pay attention to your surroundings. No texting or listening to music with headphones on – you need all your senses to stay safe.If you're on a bike, you've got some extra responsibilities. Follow all the same rules as drivers – stop at red lights and stop signs, signal your turns, and ride in the same direction as traffic. And for Pete's sake, wear a helmet! You don't want to end up with a scrambled brain, do you? Oh, and one more thing: no riding on sidewalks. That's just asking for trouble with pedestrians.For all you lucky ducks who get to ride in cars, your job is to be a super awesome role model for your parents or whoever's driving. That means always wearing your seatbelt (no exceptions!) and behaving like a respectful little angel. No distracting thedriver with your antics or sibling squabbles. And if you see the driver doing something unsafe, like texting or speeding, speak up! They'll thank you later for keeping them in line.But traffic safety isn't just about following rules – it's also about being a considerate human being. That means not jaywalking or playing in the street (duh), but it also means things like not littering from your car and not blaring your music super loud when you're driving around. We all share the roads, so let's keep them clean and peaceful for everyone.I know this all might sound like a lot of rules and nagging, but trust me, it's for your own good. The roads can be a dangerous place if we're not careful. By following these tips, we can all do our part to keep our community safe and avoid those tragic accidents.So there you have it, my fellow kids – the ultimate guide to traffic safety from one of your own. Let's lead by example and show the grown-ups how it's done! Safe travels, and don't forget to look both ways before you cross the street. Peace out!篇6Keeping Safe on the Roads and StreetsHi there! I'm just a regular 7th grader but I wanted to share some thoughts on being safe when you're walking, biking, or riding in a vehicle. Traffic safety is super important and can literally save lives.First up, let's talk about walking safely. Whenever I'm out walking, especially if I'm by myself, I always use sidewalks or paths away from the road when I can. If there aren't sidewalks, I walk facing traffic so I can see any cars coming. I also avoid distractions like looking at my phone so I can pay full attention. My parents taught me to cross streets at corners or crosswalks after looking left, right, then left again before going. If cars are stopped, I make sure to watch for any that might not see me crossing. Oh, and a biggie - wear bright or reflective clothes if it's dark out so drivers can see you better.When I'm biking, there are some extra safety steps. I always wear a properly fitted helmet - it can prevent serious head injuries if I crash. Riding predictably in the same direction as traffic is a must, obeying street signs and signals just like cars do. Using hand signals for turns helps drivers anticipate where I'm going. Staying focused and riding defensively, assuming drivers may not see me, is key. Things like potholes, drain grates, ordebris can cause crashes too, so I watch out for those hazards. At night, I have front and rear bicycle lights to be visible.Sometimes I'm a passenger when my parents or other trusted adults drive. In those cases, I buckle up with a seat belt or use a booster seat every single trip, no matter how short it is. Sitting properly and avoiding distracting the driver is important. If I think the driver is driving unsafely, like speeding or using their phone, I'll speak up about it calmly.Believe it or not, even kids can help keep drivers alert by our actions. If I'm playing near a road, I stay far away from the edges where a car could potentially go off the roadway. When I'm old enough to start driving someday, I'll definitely put these safety habits to good use behind the wheel too.There are all kinds of traffic out there - pedestrians, bicyclists, cars, trucks, buses, motorcycles, and more. Everyone has to share the roads and watch out for each other. A momentary distraction or bad decision could have devastating consequences. Following the "rules of the road" as a walker or bicyclist, and being aware of your surroundings, makes the streets safer for everyone. Let's all do our part!。
★阅读难点关键句200句（以包括译文）★1. Wearing a seat belt saves lives; it reduces your chance of death or serious injury by more than half.1、系好安全带能够挽救性命，它能将丧生和重伤的概率减少一半以上。
2. But it will be the driver‘s responsibility to make sure that children under 14 do not ride i n the front unless they are wearing a seat belt of some kind.2、但是司机有责任确保14岁以下的孩子不要坐在前排，除非他们系好了安全带。
3. However, you do not have to wear a seat belt if you are reversing your vehicle; or you are making a localdelivery or collection using a special vehicle; or if you have a valid medical certificate which excuses you from wearing it.3、当然，如果有以下情况你可以不系安全带：你在倒车时，或者你用一种特殊交通工具进行当地的货物运送、收集时，或者你有合法的医学证明你不能系安全带时。
4. Remember you may be taken to court for not doing so, and you may be fined if you cannot prove to the court that you have been excused from wearing it.4、注意你如果不这么做（系安全带）的话，你有可能被告上法庭，而且你有可能被处以罚款除非你能证明你有不带安全带的理由。
哈师大附中2024年高三第三次模拟考试英语试卷注意事项：1．答卷前，考生务必将自己的姓名、准考证号填写在答题卡上。
2．答选择题时，选出每小题答案后，用铅笔把答题卡对应题目的答案标号涂黑。
如需改动，用橡皮擦干净后，再选涂其他答案标号。
答非选择题时，将答案写在答题卡上。
写在本试卷上无效。
3．考试结束后，将本试卷和答题卡一并交回。
第一部分听力（共两节，满分30分）第一节（共5小题；每小题1.5分，满分7.5分）听下面5段对话。
每段对话后有一个小题，从题中所给的A、B、C三个选项中选出最佳选项。
听完每段对话后，你都有10秒钟的时间来回答有关小题和阅读下一小题。
每段对话仅读一遍。
例：How much is the shirt?A．£ 19.15. B．£9.18. C．£9.15.答案是C。
1．What is Saratoga well known for?A．Its natural scenery. B．Its various races. C．Its fast horses.2．Where is the butter?A．In the bowl. B．In the fridge. C．In the cupboard.3．Which programme does the girl want to watch?A．Italian gardens. B．A dance competition. C．A history programme.4．What does the man mean?A．He got on the wrong bus.B．He has to wait for the bus.C．He will be late for his flight.5．What are the speakers discussing?A．A hotel room. B．The man’s family.C．A reasonable offer.第二节（共15小题，每小题1.5分，满分22.5分）听下面5段对话或独白。
The Trial That Rocked the WorldJohn ScopesA buzz ran through the crowd as I took my place in the packed court on that sweltering July day in 1925. The counsel for my defence was the famous criminal lawyer Clarence Darrow. Leading counsel for the prosecution was William Jennings Bryan, the silver-tongued orator , three times Democratic nominee for President of the United States, and leader of the fundamentalist movement that had brought about my trial.在一九二五年七月的那个酷热日子里，当我在挤得水泄不通的法庭里就位时，人群中响起一阵嘁嘁喳喳的议论声。
我的辩护人是著名刑事辩护律师克拉伦斯•达罗。
担任主控官的则是能说会道的演说家威廉•詹宁斯•布莱恩，他曾三次被民主党提名为美国总统候选人，而且还是导致我这次受审的基督教原教旨主义运动的领导人。
A few weeks before I had been an unknown school-teacher in Dayton, a little town in the mountains of Tennessee. Now I was involved in a trial reported the world over. Seated in court, ready to testify on my behalf, were a dozen distinguished professors and scientists, led by Professor Kirtley Mather of Harvard University. More than 100 reporters were on hand, and even radio announcers, who for the first time in history were to broadcast a jury trial. "Don't worry, son, we'll show them a few tricks," Darrow had whispered throwing a reassuring arm round my shoulder as we were waiting for the court to open. 几个星期之前，我还只是田纳西州山区小镇戴顿的一名默默无闻的中学教员，而现在我却成了一次举世瞩目的庭审活动的当事人。
肯尼迪就职演讲观后感篇一：肯尼迪就职演说评析美国第三十五任总统JohnFitzgeraldFrancisKennedy(1917-1963)约翰.弗.肯尼迪1961年元月20日在首都华盛顿国会大厦前发表“就职演说”时，我在读初中三年级，学的是俄语。
直到1980年，我才在美国出版的“EnglishForToday”“今日英语”教材的第五册里阅读到了这篇演说，而且还听了这篇演说的实况录音。
现在这篇演说已被一字未删地选入《advancedEnglish》“高级英语”（张汉熙主编，商务印书馆出版发行），《21centurycollegeEnglish》“二十一世纪大学英语”（复旦大学，交通大学主编;高等教育出版社，复旦大学出版社出版发行）英语教材里作为高等院校的英语学习教材。
1980年，那时大学外语教学还是很原始落后的。
我想得到英语版的联合国“人权宣言”，但在当时武汉的中南财经学院图书馆里没有。
找到武汉大学图书馆，那里才只有一本油印的“人权宣言”小册子。
我想得到英文版的“中华人民共和国刑法”这书，武汉的外文书店买不到。
我托原北京地院外语老师去北京外国专家局找有关专家打听此书，专家说，《刑法》英文译文由他翻译，正在他手里，由于没有出版，他不能外借。
肯尼迪“就职演说”是在演说之后十九年被我们看到。
时过境迁，20XX年元月20日，全世界几乎所有的人都能从网上及各种媒体上听到，见到，读到美国第一位黑人总统奥巴马的“就职演说”。
虽然有的人看到的是被有些媒体屏掉了（RecallthatearliergenerationsfaceddownFascismandcommunismnotjust withmissilesandtanksbutwithsturdyalliancesandenduringconvictions.我们在此回忆先辈，他们战胜了法西斯主义和共产主义，靠的不仅是导弹，坦克；更是靠坚定的盟友和不移的信念。
手惰市安逸阳光实验学校提纲作文(2013·安徽卷)假设你校英语社团举办以“讲求文明,从我做起”为主题的征文活动,请你以“On the Way to School”为题,写一篇英语短文。
内容主要包括:1.遵守交通法规;2.注意举止文明。
注意:1.词数120左右;2.可以适当增加细节,以使行文连贯;3.短文中不能出现与本人相关的信息;4.短文的题目已为你拟好,不计入总词数On the Way to School【参考范文】On the Way to SchoolThese days, breaking traffic rules and littering are not uncommon, causing serious harm to life and the environment. Changing this situation requires considerable effort on the part of everyone. As for me, it should start on my way to school.I will keep traffic rules in mind all the way. If I ride a bike, I’ll always keep to the right and never cross a road until the traffic light turns green. If I walk, I’ll never forget to use the pedestrian crossing. Meanwhile, I will regard it as my duty to help keep our environment clean and healthy. Not only will I keep from littering and spitting anywhere.I will also help clean up the roadside litter whenever possible. I hope my behavior will make a difference.(2013·山东卷)第二节:写作(满分30分)假设你是新华中学的学生李华, 你的朋友Tom一周前给你发电子邮件, 询问你暑假里的打算, 但你因准备期末考试未能及时回复。
初三英语作文交通安全范文初三英语作文“交通安全人人有责”Obeying the Traffic Laws「遵守交通规则」I am often very afraid to cross large wide streets. I always go to the traffic light and use the crosswalk, but many times I have been frightened. When the light changes to green, I still need to look both directions to check the traffic On many oasions a speeding motorcycle or bicycle or once a truck drove past the red light and across the pedestrian's When I have my bicycle, I get off and walk across the street, but always someone crosses the red light. Once at the intersection near National Taiwan University I saw an aident: a taxi had stopped for the light, and another truck came from behind and did not stop. For safety, it is very important for everyone to obey the traffic laws.另附：Traffic Safety（交通安全）Traffic safety is everybody's business. Records showthat every year a lot of people die in traffic aidents. Some of the aidents are due to mechanical problems. However, most of them are the results of careless and reckless driving, and could be avoided. A lot of people disregard traffic signals and rules. They drive regardless of speed limits, run through red lights, drive in the wrong direction, talk and laugh while driving, and turn as they wish without giving signals. They don't slow down while approaching crossroads. So many people violate traffic regulations that we cannot put too much emphasis on the importance of traffic safety. Only when everybody thinks traffic safety is everybody's business can we be safe driving on roads and walking on sidewalks.交通安全人人有责。
无人驾驶的好处和潜在危险英语作文全文共3篇示例，供读者参考篇1The Amazing Self-Driving Cars of the Future!Have you ever dreamed of riding in a car that can drive itself? Well, that dream is becoming a reality with the development of self-driving cars! These incredible vehicles use special sensors, cameras, and computers to navigate the roads without a human driver. Imagine how convenient it would be to hop into aself-driving car, tell it where you want to go, and then sit back and relax while it takes you there safely. No more stressful traffic jams or getting lost! Self-driving cars are like having your own personal chauffeur.There are many awesome benefits to self-driving cars that make them really exciting. One of the biggest advantages is increased safety on the roads. Most car accidents are caused by human error, like distracted driving, speeding, or driving under the influence. But self-driving cars don't get tired, distracted, or make silly mistakes like humans do. Their sensors and computerscan react much faster than any person, helping to avoid accidents and keep everyone safe.Self-driving cars can also be a huge help for people who can't drive themselves, like the elderly or those with disabilities. Instead of relying on others for rides, they could simply call for a self-driving car to pick them up and take them wherever they need to go, giving them more independence and freedom.Another great benefit is that self-driving cars can help reduce traffic and pollution. Since they can communicate with each other and drive more efficiently, they can reduce the number of cars on the road and traffic jams. This means less time wasted sitting in traffic and less harmful emissions being released into the air we breathe. Isn't that awesome?While self-driving cars sound really cool, there are also some potential dangers we need to be aware of. One big concern is cyber security. Since these cars rely so heavily on computers and networks, they could potentially be hacked by bad people trying to cause trouble. Imagine if a hacker took control of yourself-driving car and made it go somewhere you didn't want to go! That would be really scary.Another worry is what happens if the sensors or computers fail while the car is driving. Without a human ready to take over,a system failure could lead to a serious accident. The technology needs to be extremely reliable and have good backup systems in place.There are also ethical questions about how self-driving cars should be programmed to deal with difficult situations. For example, if a self-driving car had to choose between hitting a group of people or swerving into a wall and potentially hurting its passengers, what should it be programmed to do? These are tough decisions that need to be carefully thought through.Despite these potential dangers, many experts believe that the benefits of self-driving cars outweigh the risks, especially as the technology continues to improve and become more reliable. With proper safety measures, cyber security precautions, and clear programming guidelines, self-driving cars could make our roads significantly safer and more efficient.Just imagine how cool it would be to tell your self-driving car, "Take me to the park, please!" and have it drive you there while you play video games or read a book. Or think about how much easier it would be for your parents or grandparents to get around if they had a self-driving car at their service. The future of transportation is looking incredibly exciting and convenient.Of course, we'll still need to be cautious and make sureself-driving cars are as safe and secure as possible. But with continued research and development, these amazing vehicles could revolutionize the way we travel and make our lives much easier. Who knows, maybe one day you'll be telling your own self-driving car where to take you on your next adventure! The possibilities are endless, and the future of self-driving cars is looking brighter than ever.篇2The Future is Self-Driving Cars!Robots and computers are becoming a bigger part of our lives every day. One new technology that could change how we get around is self-driving cars! Instead of a person driving the car, it would be controlled by a computer program. This might sound crazy, but companies like Google, Tesla, and others are already testing self-driving car prototypes on the roads today. Let me tell you about some of the awesome benefits of self-driving cars, but also some of the dangers we need to be careful about.Benefits of Self-Driving CarsNo More Distracted DrivingOne of the biggest benefits of self-driving cars is there would be way fewer car accidents. So many crashes today happen because people get distracted while driving. They might be texting, eating, putting on makeup, or just not paying full attention to the road. The computer driving the self-driving car would never get distracted or lose focus. Its sensors and cameras would constantly be monitoring the surroundings, brake instantly if a kid runs into the street, and take the safest driving route every time.Independence for People Who Can't DriveAnother huge plus of self-driving cars is that many people who aren't able to drive today could get around easily. This includes kids like me, but also people who are blind or have other disabilities that make driving difficult. Elderly people whose driving skills aren't as good anymore could also getself-driving cars to stay independent instead of having to ask others for rides. With a self-driving car, we could just tell it where to go and it would take us there safely!Less Traffic and PollutionDid you know that a lot of traffic is caused by human drivers not being good at things like merging, keeping a consistent speed, and other driving tasks? A fleet of self-driving cars thatare all communicating with each other could flow smoothly in and out of traffic, maximizing the space on the roads. This would cut down hugely on the time people spend stuck in traffic jams. Since self-driving cars could also drive much more efficiently than humans and accelerate/brake smoothly, it's estimated they could reduce energy consumption by a lot. Less fuel being burned by cars means less pollution too, which is better for the environment.You Can Do Other StuffOne of the coolest benefits for kids like me is that you wouldn't actually have to spend any time driving yourself if you had a self-driving car! Instead of your parents having to pull their attention away from the road, self-driving cars would free them up to keep parents entertained on road trips. Parents could also use that time to answer work emails, study, or read a book since they wouldn't need to focus on driving at all. Some day when I'm older and have my own self-driving car, I could even play video games or do homework while it takes me wherever I need to go!Potential Dangers of Self-Driving CarsWhile self-driving cars offer some amazing benefits, there are also some dangers and risks we would need to be really careful about:Hacking RisksSince self-driving cars would be controlled by computer programs, there is a risk that hackers could potentially take control of the car's systems and cause accidents or chaos. Companies would need to make sure their self-driving car software has incredibly strong security protection.Software Bugs/ErrorsJust like sometimes apps or video games have glitches due to bugs in the coding, it's possible there could be errors in the complex software running self-driving cars. If not properly tested, these could cause the cars to make mistakes and put people's safety at risk. Software for self-driving cars would need to go through exhaustive checks.Weather/Sensor IssuesThe sensors and cameras that allow self-driving cars to "see" the world around them could potentially get blinded or malfunction in certain weather conditions like heavy rain, fog, or snow. Engineers would have to make sure the sensor systems are robust enough to handle any type of road conditions.Jumping to New Technology Too QuicklyAs revolutionary as self-driving cars aim to be, switching over too quickly before the technology is proven to be 100% safe could lead to disasters. That's why self-driving cars are being slowly phased onto roads and tested for years to identify all possible risks before they become widespread.Job LossesOne downside of self-driving cars is that they could put a lot of professional drivers out of jobs, like truck drivers, taxi/Uber drivers, and bus drivers. While this would be bad for those workers, hopefully new jobs could be created in other areas to make up for it.CostFinally, ensuring self-driving cars can operate safely under any conditions while having backup systems is very complex and costly. The first self-driving cars that come out might only be affordable for wealthy people, not the average family. But over time as production increases, the costs could come down.My Thoughts on Self-Driving CarsOverall, I'm really excited about the future potential ofself-driving cars! Being able to go anywhere just by telling my car's computer where I want to go sounds incredibly convenient.But safety has to be the top priority before they become mainstream on the roads. Maybe by the time I'm old enough to drive, self-driving cars will finally be the norm and households won't need multiple cars since one self-driving car could dispersely shuttle each family member around on their schedules. Just one of the many ways this new technology could transform transportation!篇3The Awesome and Scary World of Self-Driving CarsSelf-driving cars are vehicles that can drive themselves without a human driver! They use sensors, cameras, and computers to see the road, follow traffic laws, and get you where you need to go. Self-driving cars are an amazing new technology that could make driving way more convenient and safer. But they also have some potential dangers we need to think about. Let me tell you about the awesome upsides and scary downsides of these futuristic vehicles!The Benefits of Self-Driving CarsNot having to drive sounds like a kid's dream come true! With a self-driving car, you could kick back, play video games, watch movies, eat snacks, or even take a nap during your trip. Nomore keeping your eyes glued to the road. The car's computers and sensors will do all the hard work for you.Self-driving cars could also give freedom to people who can't drive normal cars due to age or disabilities. Little kids, elderly folks, and people with vision problems or other challenges could all get around independently in a self-driving vehicle. How cool is that?The coolest benefit of all might be the potential for far fewer car accidents. Human drivers often get distracted, drive recklessly, or just make mistakes that cause crashes. Butself-driving cars have lightning-fast sensors and never get sleepy or distracted. They are designed to obey all traffic laws and drive more safely than humans. Imagine how many lives could be saved!Potential Dangers of Self-Driving CarsAs awesome as self-driving cars sound, they also have some pretty scary potential risks we need to think about. One of the biggest worries is equipment failure or software glitches. What if the sensors or computers on a self-driving car malfunction while driving? A minor glitch could cause a dangerous situation or even a terrible accident.Hackers are also a concern for any computer-controlled system like self-driving cars. Bad people could potentially hack into the systems and cause chaos - like suddenly taking control of the vehicle against your will. That's a chilling thought!Another issue is the question of responsibility if aself-driving car does get into an accident. Is it the car company's fault for any crashes and injuries? Or does some responsibility fall on the person riding in the self-driving car? These types of legal issues will need to be sorted out.Self-driving cars also bring up some tricky ethical dilemmas. If an accident is unavoidable, how should the car's programming decide between two bad options? Should it risk injuring pedestrians or put its passengers in harm's way? There are no easy answers.Losing Driving FreedomFinally, some people are worried that self-driving cars could eventually make human driving illegal, at least in certain areas. After all, if autonomous vehicles are vastly safer than human drivers, why would we allow the unsafe human option? While this could prevent many accidents, it would also take away the freedom and feeling of control that some people love about driving themselves.Brave New WorldSo those are some of the major pros and cons to think about with self-driving cars. I'm still not sure if I'm more excited about their potential awesomeness or scared of the risks. Self-driving vehicles could make driving incredibly easy, convenient, and safe. But their autonomy and reliance on complex technology also creates new types of dangers. I guess only time will tell if they turn out to be a fantastic innovation or a troubling can of worms.Either way, the world is going to look very different in the future with these computer-chauffeured cars sharing the roads. As a kid, I find it all quite fascinating and futuristic. Self-driving cars feel like science fiction brought to life! Part of me will be sad to see traditional human driving go away. But mostly, I'm excited to grow up in such an amazingly advanced world. Just don't be surprised if you see me playing video games from the backseat of the family's new self-driving car!。
from Solving Pedestrian Detection?Mohamed Omran,Jan Hosang,and Bernt SchielePlanck Institute for Informatics Saarbrücken,Germanystname@mpi-inf.mpg.deAbstractEncouraged by the recent progress in pedestrian detec-tion,we investigate the gap between current state-of-the-art methods and the “perfect single frame detector”.We en-able our analysis by creating a human baseline for pedes-trian detection (over the Caltech dataset),and by manually clustering the recurrent errors of a top detector.Our res-ults characterize both localization and background-versus-foreground errors.To address localization errors we study the impact of training annotation noise on the detector performance,and show that we can improve even with a small portion of sanitized training data.To address background/foreground discrimination,we study convnets for pedestrian detection,and discuss which factors affect their performance.Other than our in-depth analysis,we report top perform-ance on the Caltech dataset,and provide a new sanitized set of training and test annotations 1.1.IntroductionObject detection has received great attention during re-cent years.Pedestrian detection is a canonical sub-problem that remains a popular topic of research due to its diverse applications.Despite the extensive research on pedestrian detection,recent papers still show significant improvements,suggest-ing that a saturation point has not yet been reached.In this paper we analyse the gap between the state of the art and a newly created human baseline (section 3.1).The results indicate that there is still a ten fold improvement to be made before reaching human performance.We aim to investigate which factors will help close this gap.We analyse failure cases of top performing pedestrian detectors and diagnose what should be changed to further push performance.We show several different analysis,in-cluding human inspection,automated analysis of problem1Ifyou are interested in our new annotations,please contact Shanshan Zhang.1010101010Figure 1:Overview of the top results on the Caltech-USA pedestrian benchmark (CVPR2015snapshot).At ∼95%recall,state-of-the-art detectors make ten times more errors than the human baseline.cases (e.g.blur,contrast),and oracle experiments (section 3.2).Our results indicate that localization is an important source of high confidence false positives.We address this aspect by improving the training set alignment quality,both by manually sanitising the Caltech training annotations and via algorithmic means for the remaining training samples (sections 3.3and 4.1).To address background versus foreground discrimina-tion,we study convnets for pedestrian detection,and dis-cuss which factors affect their performance (section 4.2).1.1.Related workIn the last years,diverse efforts have been made to im-prove the performance of pedestrian detection.Following the success of integral channel feature detector (ICF)[6,5],many variants [22,23,16,18]were proposed and showed significant improvement.A recent review of pedestrian de-tection [3]concludes that improved features have been driv-ing performance and are likely to continue doing so.It also shows that optical flow [19]and context information [17]are complementary to image features and can further boost 1a r X i v :1602.01237v 1 [c s .C V ] 3 F eb 2016detection accuracy.Byfine-tuning a model pre-trained on external data convolution neural networks(convnets)have also reached state-of-the-art performance[15,20].Most of the recent papers focus on introducing novelty and better results,but neglect the analysis of the resulting system.Some analysis work can be found for general ob-ject detection[1,14];in contrast,in thefield of pedestrian detection,this kind of analysis is rarely done.In2008,[21] provided a failure analysis on the INRIA dataset,which is relatively small.The best method considered in the2012 Caltech dataset survey[7]had10×more false positives at20%recall than the methods considered here,and no method had reached the95%mark.Since pedestrian detection has improved significantly in recent years,a deeper and more comprehensive analysis based on state-of-the-art detectors is valuable to provide better understanding as to where future efforts would best be invested.1.2.ContributionsOur key contributions are as follows:(a)We provide a detailed analysis of a state-of-the-art ped-estrian detection system,providing insights into failure cases.(b)We provide a human baseline for the Caltech Pedestrian Benchmark;as well as a sanitised version of the annotations to serve as new,high quality ground truth for the training and test sets of the benchmark.The data will be public. (c)We analyse how much the quality of training data affects the detector.More specifically we quantify how much bet-ter alignment and fewer annotation mistakes can improve performance.(d)Using the insights of the analysis,we explore variants of top performing methods:filtered channel feature detector [23]and R-CNN detector[13,15],and show improvements over the baselines.2.PreliminariesBefore delving into our analysis,let us describe the data-sets in use,their metrics,and our baseline detector.2.1.Caltech-USA pedestrian detection benchmarkAmongst existing pedestrian datasets[4,9,8],KITTI [11]and Caltech-USA are currently the most popular ones. In this work we focus on the Caltech-USA benchmark[7] which consists of2.5hours of30Hz video recorded from a vehicle traversing the streets of Los Angeles,USA.The video annotations amount to a total of350000bound-ing boxes covering∼2300unique pedestrians.Detec-tion methods are evaluated on a test set consisting of4024 frames.The provided evaluation toolbox generates plotsFilter type MR O−2ACF[5]44.2SCF[3]34.8LDCF[16]24.8RotatedFilters19.2Checkerboards18.5Table1:Thefiltertype determines theICF methods quality.Base detector MR O−2+Context+FlowOrig.2Ped[17]48~5pp/Orig.SDt[19]45/8ppSCF[3]355pp4ppCheckerboards19~01ppTable2:Detection quality gain ofadding context[17]and opticalflow[19],as function of the base detector.for different subsets of the test set based on annotation size, occlusion level and aspect ratio.The established proced-ure for training is to use every30th video frame which res-ults in a total of4250frames with∼1600pedestrian cut-outs.More recently,methods which can leverage more data for training have resorted to afiner sampling of the videos [16,23],yielding up to10×as much data for training than the standard“1×”setting.MR O,MR N In the standard Caltech evaluation[7]the miss rate(MR)is averaged over the low precision range of [10−2,100]FPPI.This metric does not reflect well improve-ments in localization errors(lowest FPPI range).Aiming for a more complete evaluation,we extend the evaluation FPPI range from traditional[10−2,100]to[10−4,100],we denote these MR O−2and MR O−4.O stands for“original an-notations”.In section3.3we introduce new annotations, and mark evaluations done there as MR N−2and MR N−4.We expect the MR−4metric to become more important as de-tectors get stronger.2.2.Filtered channel features detectorFor the analysis in this paper we consider all methods published on the Caltech Pedestrian benchmark,up to the last major conference(CVPR2015).As shown infigure1, the best method at the time is Checkerboards,and most of the top performing methods are of its same family.The Checkerboards detector[23]is a generalization of the Integral Channels Feature detector(ICF)[6],which filters the HOG+LUV feature channels before feeding them into a boosted decision forest.We compare the performance of several detectors from the ICF family in table1,where we can see a big improve-ment from44.2%to18.5%MR O−2by introducingfilters over the feature channels and optimizing thefilter bank.Current top performing convnets methods[15,20]are sensitive to the underlying detection proposals,thus wefirst focus on the proposals by optimizing thefiltered channel feature detectors(more on convnets in section4.2). Rotatedfilters For the experiments involving train-ing new models(in section 4.1)we use our own re-implementation of Checkerboards[23],based on the LDCF[16]codebase.To improve the training time we decrease the number offilters from61in the originalCheckerboards down to9filters.Our so-called Rota-tedFilters are a simplified version of LDCF,applied at three different scales(in the same spirit as Squares-ChnFtrs(SCF)[3]).More details on thefilters are given in the supplementary material.As shown in table1,Ro-tatedFilters are significantly better than the original LDCF,and only1pp(percent point)worse than Checker-boards,yet run6×faster at train and test time. Additional cues The review[3]showed that context and opticalflow information can help improve detections. However,as the detector quality improves(table1)the re-turns obtained from these additional cues erodes(table2). Without re-engineering such cues,gains in detection must come from the core detector.3.Analysing the state of the artIn this section we estimate a lower bound on the re-maining progress available,analyse the mistakes of current pedestrian detectors,and propose new annotations to better measure future progress.3.1.Are we reaching saturation?Progress on pedestrian detection has been showing no sign of slowing in recent years[23,20,3],despite recent im-pressive gains in performance.How much progress can still be expected on current benchmarks?To answer this ques-tion,we propose to use a human baseline as lower bound. We asked domain experts to manually“detect”pedestrians in the Caltech-USA test set;machine detection algorithms should be able to at least reach human performance and, eventually,superhuman performance.Human baseline protocol To ensure a fair comparison with existing detectors,we focus on the single frame mon-ocular detection setting.Frames are presented to annotators in random order,and without access to surrounding frames from the source videos.Annotators have to rely on pedes-trian appearance and single-frame context rather than(long-term)motion cues.The Caltech benchmark normalizes the aspect ratio of all detection boxes[7].Thus our human annotations are done by drawing a line from the top of the head to the point between both feet.A bounding box is then automatically generated such that its centre coincides with the centre point of the manually-drawn axis,see illustration infigure2.This procedure ensures the box is well centred on the subject (which is hard to achieve when marking a bounding box).To check for consistency among the two annotators,we produced duplicate annotations for a subset of the test im-ages(∼10%),and evaluated these separately.With a Intersection over Union(IoU)≥0.5matching criterion, the results were identical up to a single boundingbox.Figure2:Illustration of bounding box generation for human baseline.The annotator only needs to draw a line from the top of the head to the central point between both feet,a tight bounding box is then automatically generated. Conclusion Infigure3,we compare our human baseline with other top performing methods on different subsets of the test data(varying height ranges and occlu-sion levels).Wefind that the human baseline widely out-performs state-of-the-art detectors in all settings2,indicat-ing that there is still room for improvement for automatic methods.3.2.Failure analysisSince there is room to grow for existing detectors,one might want to know:when do they fail?In this section we analyse detection mistakes of Checkerboards,which obtains top performance on most subsets of the test set(see figure3).Since most top methods offigure1are of the ICF family,we expect a similar behaviour for them too.Meth-ods using convnets with proposals based on ICF detectors will also be affected.3.2.1Error sourcesThere are two types of errors a detector can do:false pos-itives(detections on background or poorly localized detec-tions)and false negatives(low-scoring or missing pedes-trian detections).In this analysis,we look into false positive and false negative detections at0.1false positives per im-age(FPPI,1false positive every10images),and manually cluster them(one to one mapping)into visually distinctive groups.A total of402false positive and148false negative detections(missing recall)are categorized by error type. False positives After inspection,we end up having all false positives clustered in eleven categories,shown infig-ure4a.These categories fall into three groups:localization, background,and annotation errors.Background errors are the most common ones,mainly ver-tical structures(e.g.figure5b),tree leaves,and traffic lights. This indicates that the detectors need to be extended with a better vertical context,providing visibility over larger struc-tures and a rough height estimate.Localization errors are dominated by double detections2Except for IoU≥0.8.This is due to issues with the ground truth, discussed in section3.3.Reasonable (IoU >= 0.5)Height > 80Height in [50,80]Height in [30,50]020406080100HumanBaselineCheckerboards RotatedFiltersm i s s r a t eFigure 3:Detection quality (log-average miss rate)for different test set subsets.Each group shows the human baseline,the Checkerboards [23]and RotatedFilters detectors,as well as the next top three (unspecified)methods (different for each setting).The corresponding curves are provided in the supplementary material.(high scoring detections covering the same pedestrian,e.g.figure 5a ).This indicates that improved detectors need to have more localized responses (peakier score maps)and/or a different non-maxima suppression strategy.In sections 3.3and 4.1we explore how to improve the detector localiz-ation.The annotation errors are mainly missing ignore regions,and a few missing person annotations.In section 3.3we revisit the Caltech annotations.False negatives Our clustering results in figure 4b show the well known difficulty of detecting small and oc-cluded objects.We hypothesise that low scoring side-view persons and cyclists may be due to a dataset bias,i.e.these cases are under-represented in the training set (most per-sons are non-cyclist walking on the side-walk,parallel to the car).Augmenting the training set with external images for these cases might be an effective strategy.To understand better the issue with small pedestrians,we measure size,blur,and contrast for each (true or false)de-tection.We observed that small persons are commonly sat-urated (over or under exposed)and blurry,and thus hypo-thesised that this might be an underlying factor for weak detection (other than simply having fewer pixels to make the decision).Our results indicate however that this is not the case.As figure 4c illustrates,there seems to be no cor-relation between low detection score and low contrast.This also holds for the blur case,detailed plots are in the sup-plementary material.We conclude that the small number of pixels is the true source of difficulty.Improving small objects detection thus need to rely on making proper use of all pixels available,both inside the window and in the surrounding context,as well as across time.Conclusion Our analysis shows that false positive er-rors have well defined sources that can be specifically tar-geted with the strategies suggested above.A fraction of the false negatives are also addressable,albeit the small and oc-cluded pedestrians remain a (hard and)significant problem.20406080100120# e r r o r s 0100200300loc a liz a tion ba c k g round a nnota e rrors#e r r o r s (a)False positive sources15304560# e r r o r s (b)False negative sources(c)Contrast versus detection scoreFigure 4:Errors analysis of Checkerboards [23]on the test set.(a)double detectionFigure 5:Example of analysed false positive cases (red box).Additional ones in supplementary material.3.2.2Oracle test casesThe analysis of section 3.2.1focused on errors counts.For area-under-the-curve metrics,such astheones used in Caltech,high-scoring errors matter more than low-scoring ones.In this section we directly measure the impact of loc-alization and background-vs-foreground errors on the de-tection quality metric (log-average miss-rate)by using or-acle test cases.In the oracle case for localization,all false positives that overlap with ground truth are ignored for evaluation.In the oracle tests for background-vs-foreground,all false posit-ives that do not overlap with ground truth are ignored.Figure 6a shows that fixing localization mistakes im-proves performance in the low FPPI region;while fixing background mistakes improves results in the high FPPI re-gion.Fixing both types of mistakes results zero errors,even though this is not immediately visible due to the double log plot.In figure 6b we show the gains to be obtained in MR O −4terms by fixing localization or background issues.When comparing the eight top performing methods we find that most methods would boost performance significantly by fix-ing either problem.Note that due to the log-log nature of the numbers,the sum of localization and background deltas do not add up to the total miss-rate.Conclusion For most top performing methods localiz-ation and background-vs-foreground errors have equal im-pact on the detection quality.They are equally important.3.3.Improved Caltech-USA annotationsWhen evaluating our human baseline (and other meth-ods)with a strict IoU ≥0.8we notice in figure 3that the performance drops.The original annotation protocol is based on interpolating sparse annotations across multiple frames [7],and these sparse annotations are not necessar-ily located on the evaluated frames.After close inspection we notice that this interpolation generates a systematic off-set in the annotations.Humans walk with a natural up and down oscillation that is not modelled by the linear interpol-ation used,thus in most frames have shifted bounding box annotations.This effect is not noticeable when using the forgiving IoU ≥0.5,however such noise in the annotations is a hurdle when aiming to improve object localization.1010−210−110010false positives per image18.47(33.20)% Checkerboards15.94(25.49)% Checkerboards (localization oracle)11.92(26.17)% Checkerboards (background oracle)(a)Original and two oracle curves for Checkerboards de-tector.Legend indicates MR O −2 MR O −4 .(b)Comparison of miss-rate gain (∆MR O −4)for top performing methods.Figure 6:Oracle cases evaluation over Caltech test set.Both localization and background-versus-foreground show important room for improvement.(a)False annotations (b)Poor alignmentFigure 7:Examples of errors in original annotations.New annotations in green,original ones in red.This localization issues together with the annotation er-rors detected in section 3.2.1motivated us to create a new set of improved annotations for the Caltech pedestrians dataset.Our aim is two fold;on one side we want to provide a more accurate evaluation of the state of the art,in particu-lar an evaluation suitable to close the “last 20%”of the prob-lem.On the other side,we want to have training annotations and evaluate how much improved annotations lead to better detections.We evaluate this second aspect in section 4.1.New annotation protocol Our human baseline focused on a fair comparison with single frame methods.Our new annotations are done both on the test and training 1×set,and focus on high quality.The annotators are allowed to look at the full video to decide if a person is present or not,they are request to mark ignore regions in areas cov-ering crowds,human shapes that are not persons (posters,statues,etc.),and in areas that could not be decided as cer-tainly not containing a person.Each person annotation is done by drawing a line from the top of the head to the point between both feet,the same as human baseline.The annot-ators must hallucinate head and feet if these are not visible. When the person is not fully visible,they must also annotate a rectangle around the largest visible region.This allows to estimate the occlusion level in a similar fashion as the ori-ginal Caltech annotations.The new annotations do share some bounding boxes with the human baseline(when no correction was needed),thus the human baseline cannot be used to do analysis across different IoU thresholds over the new test set.In summary,our new annotations differ from the human baseline in the following aspects:both training and test sets are annotated,ignore regions and occlusions are also an-notated,full video data is used for decision,and multiple revisions of the same image are allowed.After creating a full independent set of annotations,we con-solidated the new annotations by cross-validating with the old annotations.Any correct old annotation not accounted for in the new set,was added too.Our new annotations correct several types of errors in the existing annotations,such as misalignments(figure 7b),missing annotations(false negatives),false annotations (false positives,figure7a),and the inconsistent use of“ig-nore”regions.Our new annotations will be publicly avail-able.Additional examples of“original versus new annota-tions”provided in the supplementary material,as well as visualization software to inspect them frame by frame. Better alignment In table3we show quantitative evid-ence that our new annotations are at least more precisely localized than the original ones.We summarize the align-ment quality of a detector via the median IoU between true positive detections and a give set of annotations.When evaluating with the original annotations(“median IoU O”column in table3),only the model trained with original annotations has good localization.However,when evalu-ating with the new annotations(“median IoU N”column) both the model trained on INRIA data,and on the new an-notations reach high localization accuracy.This indicates that our new annotations are indeed better aligned,just as INRIA annotations are better aligned than Caltech.Detailed IoU curves for multiple detectors are provided in the supplementary material.Section4.1describes the RotatedFilters-New10×entry.4.Improving the state of the artIn this section we leverage the insights of the analysis, to improve localization and background-versus-foreground discrimination of our baseline detector.DetectorTrainingdataMedianIoU OMedianIoU N Roerei[2]INRIA0.760.84RotatedFilters Orig.10×0.800.77RotatedFilters New10×0.760.85 Table3:Median IoU of true positives for detectors trained on different data,evaluated on original and new Caltech test.Models trained on INRIA align well with our new an-notations,confirming that they are more precise than previ-ous ones.Curves for other detectors in the supplement.Detector Anno.variant MR O−2MR N−2ACFOriginal36.9040.97Pruned36.4135.62New41.2934.33 RotatedFiltersOriginal28.6333.03Pruned23.8725.91New31.6525.74 Table4:Effects of different training annotations on detec-tion quality on validation set(1×training set).Italic num-bers have matching training and test sets.Both detectors im-prove on the original annotations,when using the“pruned”variant(see§4.1).4.1.Impact of training annotationsWith new annotations at hand we want to understand what is the impact of annotation quality on detection qual-ity.We will train ACF[5]and RotatedFilters mod-els(introduced in section2.2)using different training sets and evaluate on both original and new annotations(i.e. MR O−2,MR O−4and MR N−2,MR N−4).Note that both detect-ors are trained via boosting and thus inherently sensitive to annotation noise.Pruning benefits Table4shows results when training with original,new and pruned annotations(using a5/6+1/6 training and validation split of the full training set).As ex-pected,models trained on original/new and tested on ori-ginal/new perform better than training and testing on differ-ent annotations.To understand better what the new annota-tions bring to the table,we build a hybrid set of annotations. Pruned annotations is a mid-point that allows to decouple the effects of removing errors and improving alignment. Pruned annotations are generated by matching new and ori-ginal annotations(IoU≥0.5),marking as ignore region any original annotation absent in the new ones,and adding any new annotation absent in the original ones.From original to pruned annotations the main change is re-moving annotation errors,from pruned to new,the main change is better alignment.From table4both ACF and RotatedFilters benefit from removing annotation er-rors,even in MR O−2.This indicates that our new training setFigure 8:Examples of automatically aligned ground truth annotations.Left/right →before/after alignment.1×data 10×data aligned withMR O −2(MR O −4)MR N −2(MR N−4)Orig.Ø19.20(34.28)17.22(31.65)Orig.Orig.10×19.16(32.28)15.94(29.33)Orig.New 1/2×16.97(28.01)14.54(25.06)NewNew 1×16.77(29.76)12.96(22.20)Table 5:Detection quality of RotatedFilters on test set when using different aligned training sets.All mod-els trained with Caltech 10×,composed with different 1×+9×combinations.is better sanitized than the original one.We see in MR N −2that the stronger detector benefits more from better data,and that the largest gain in detection qual-ity comes from removing annotation errors.Alignment benefits The detectors from the ICF family benefit from training with increased training data [16,23],using 10×data is better than 1×(see section 2.1).To lever-age the 9×remaining data using the new 1×annotations we train a model over the new annotations and use this model to re-align the original annotations over the 9×portion.Be-cause the new annotations are better aligned,we expect this model to be able to recover slight position and scale errors in the original annotations.Figure 8shows example results of this process.See supplementary material for details.Table 5reports results using the automatic alignment pro-cess,and a few degraded cases:using the original 10×,self-aligning the original 10×using a model trained over original 10×,and aligning the original 10×using only a fraction of the new annotations (without replacing the 1×portion).The results indicate that using a detector model to improve overall data alignment is indeed effective,and that better aligned training data leads to better detection quality (both in MR O and MR N ).This is in line with the analysis of section 3.2.Already using a model trained on 1/2of the new annotations for alignment,leads to a stronger model than obtained when using original annotations.We name the RotatedFilters model trained using the new annotations and the aligned 9×data,Rotated-Filters-New10×.This model also reaches high me-dian true positives IoU in table 3,indicating that indeed it obtains more precise detections at test time.Conclusion Using high quality annotations for training improves the overall detection quality,thanks both to im-proved alignment and to reduced annotation errors.4.2.Convnets for pedestrian detectionThe results of section 3.2indicate that there is room for improvement by focusing on the core background versus foreground discrimination task (the “classification part of object detection”).Recent work [15,20]showed compet-itive performance with convolutional neural networks (con-vnets)for pedestrian detection.We include convnets into our analysis,and explore to what extent performance is driven by the quality of the detection proposals.AlexNet and VGG We consider two convnets.1)The AlexNet from [15],and 2)The VGG16model from [12].Both are pre-trained on ImageNet and fine-tuned over Cal-tech 10×(original annotations)using SquaresChnFtrs proposals.Both networks are based on open source,and both are instances of the R-CNN framework [13].Albeit their training/test time architectures are slightly different (R-CNN versus Fast R-CNN),we expect the result differ-ences to be dominated by their respective discriminative power (VGG16improves 8pp in mAP over AlexNet in the Pascal detection task [13]).Table 6shows that as we improve the quality of the detection proposals,AlexNet fails to provide a consistent gain,eventually worsening the results of our ICF detect-ors (similar observation done in [15]).Similarly VGG provides large gains for weaker proposals,but as the pro-posals improve,the gain from the convnet re-scoring even-tually stalls.After closer inspection of the resulting curves (see sup-plementary material),we notice that both AlexNet and VGG push background instances to lower scores,and at the same time generate a large number of high scoring false positives.The ICF detectors are able to provide high recall proposals,where false positives around the objects have low scores (see [15,supp.material,fig.9]),however convnets have difficulties giving low scores to these windows sur-rounding the true positives.In other words,despite their fine-tuning,the convnet score maps are “blurrier”than the proposal ones.We hypothesise this is an intrinsic limita-tion of the AlexNet and VGG architectures,due to their in-ternal feature pooling.Obtaining “peakier”responses from a convnet most likely will require using rather different ar-chitectures,possibly more similar to the ones used for se-mantic labelling or boundaries estimation tasks,which re-quire pixel-accurate output.Fortunately,we can compensate for the lack of spatial resolution in the convnet scoring by using bounding box regression.Adding bounding regression over VGG,and ap-plying a second round of non-maximum suppression (first NMS on the proposals,second on the regressed boxes),has。
