![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1181310%2Ff274d052ab73b080aee2e9587feffea1%2Fload-cars.png?generation=1571787431259570&alt=media\")
Welcome to the Peking University/Baidu - Autonomous Driving competition!
\n\nIn this competition, we're predicting the positions and orientations of vehicles based on images taken by a single-camera test vehicle. Please check out the Evaluation and Data pages for information about how the metric works / the data is structured.
\n\nThis should be fun / a little bit different - please let us know if you have questions!
\n\nGood luck!
", "messages": [ { "id": 671013, "postDate": "2019-11-12T06:34:52.310Z", "content": "Hi @philculliton @addisonhoward! It is looks like score calculation is needed to be described more clearly because right now we can't understand how it works. I respectfully ask you to pay attention to this discussion (https://www.kaggle.com/c/pku-autonomous-driving/discussion/116661#latest-670626) and maybe provide us a code for metrics calculation.
\n\nFor example my first submission contains all predicted cars with confidence > 0.5 and score value were 0.079, but with confidence > 0.99 score is 0.169 (4th place of leaderboard right now) by the way this submission contains only 200 cars for all images.
", "rawMarkdown": "Hi @philculliton @addisonhoward! It is looks like score calculation is needed to be described more clearly because right now we can't understand how it works. I respectfully ask you to pay attention to this discussion (https://www.kaggle.com/c/pku-autonomous-driving/discussion/116661#latest-670626) and maybe provide us a code for metrics calculation.\n\nFor example my first submission contains all predicted cars with confidence > 0.5 and score value were 0.079, but with confidence > 0.99 score is 0.169 (4th place of leaderboard right now) by the way this submission contains only 200 cars for all images.", "votes": 11, "replies": [ { "id": 671650, "postDate": "2019-11-13T01:52:40.887Z", "content": "200 cars for all images! Suppose 2000 test images has average of 10 cars per image, there will be around 20000 cars. Hence there is clearly an evaluation error for this competition. I would also suggest the competition host to provide the evaluation code.
", "rawMarkdown": "200 cars for all images! Suppose 2000 test images has average of 10 cars per image, there will be around 20000 cars. Hence there is clearly an evaluation error for this competition. I would also suggest the competition host to provide the evaluation code.", "votes": 3 } ] }, { "id": 655214, "postDate": "2019-10-22T20:13:06.873Z", "content": "Welcome to the Peking University/Baidu - Autonomous Driving competition!
\n\nIn this competition, we're predicting the positions and orientations of vehicles based on images taken by a single-camera test vehicle. Please check out the Evaluation and Data pages for information about how the metric works / the data is structured.
\n\nThis should be fun / a little bit different - please let us know if you have questions!
\n\nGood luck!
", "rawMarkdown": "Welcome to the Peking University/Baidu - Autonomous Driving competition!\n\nIn this competition, we're predicting the positions and orientations of vehicles based on images taken by a single-camera test vehicle. Please check out the Evaluation and Data pages for information about how the metric works / the data is structured.\n\nThis should be fun / a little bit different - please let us know if you have questions!\n\nGood luck!", "votes": 11 }, { "id": 655345, "postDate": "2019-10-23T00:37:46.573Z", "content": "There seem to be a lot of cars missing from the mask or detection algorithm in the training model. Three examples;
\n\n| ID | Cars detected in train.csv | Comment |\n| --- | --- | --- |\n| ID_00ab59fa6 | 13 | There's a lot more than 13 cars in this photo, even excluding those masked |\n| ID_00ac30455 | 5 | There are 2 trucks in this photo, so I'm assuming they should just be ignored completely? |\n| ID_00b7fb303 | 4 | There is also a bus in this photo. Should that just be ignored? |
\n\nOnly submit for vehicles where there is a matching 3D model?\nWhat happens if you make a submission with more vehicles detected than your algorithm detects?
", "rawMarkdown": "There seem to be a lot of cars missing from the mask or detection algorithm in the training model. Three examples;\n\n| ID | Cars detected in train.csv | Comment |\n| --- | --- | --- |\n| ID_00ab59fa6 | 13 | There's a lot more than 13 cars in this photo, even excluding those masked |\n| ID_00ac30455 | 5 | There are 2 trucks in this photo, so I'm assuming they should just be ignored completely? |\n| ID_00b7fb303 | 4 | There is also a bus in this photo. Should that just be ignored? |\n\nOnly submit for vehicles where there is a matching 3D model?\nWhat happens if you make a submission with more vehicles detected than your algorithm detects?\n", "votes": 5 }, { "id": 655332, "postDate": "2019-10-22T23:38:28.773Z", "content": "Hi @philculliton , @addisonhoward , can you please indicate libraries we have to install to load pickled files? As the sample in description is not sufficient. Thanks in advance.
\n\n
@ebouteillon @philculliton I tried with different renders including pygame and PyOpenG modules and I couldn't even open the files. However, I think we'll not be able to load and visualize models on kaggle's kernels.\nI think the hosts should upload the proper code.
", "rawMarkdown": "@ebouteillon @philculliton I tried with different renders including pygame and PyOpenG modules and I couldn't even open the files. However, I think we'll not be able to load and visualize models on kaggle's kernels.\nI think the hosts should upload the proper code." }, { "id": 655880, "postDate": "2019-10-23T15:59:24.337Z", "content": "Hi @ebouteillon @jesucristo - thanks! The pickled objects are just JSON blobs containing numpy arrays. I was able to un-pickle them on my own machines, but one of my coworkers also had a problem loading them.
\n\nIn lieu of trying to solve a bunch of pickle version issues, I've added the models as JSON files in the competition data. Let me know if that works all right, or if you still need help.
Hi @philculliton it is much better.
\n\nFor kagglers interested in seeing the models in 3D, I shared a notebook : https://www.kaggle.com/ebouteillon/load-a-3d-car-model
", "rawMarkdown": "Hi @philculliton it is much better.\n\nFor kagglers interested in seeing the models in 3D, I shared a notebook : https://www.kaggle.com/ebouteillon/load-a-3d-car-model" } ] }, { "id": 659344, "postDate": "2019-10-27T13:42:53.360Z", "content": "Thanks! Could you please double-check if distance thresholds on evaluation page are correct? I see \"Translation: 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01\" but that looks way too low, it's unlikely we'll have predictions with sub-10 cm accuracy, not talking about 1 cm?
", "rawMarkdown": "Thanks! Could you please double-check if distance thresholds on evaluation page are correct? I see \"Translation: 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01\" but that looks way too low, it's unlikely we'll have predictions with sub-10 cm accuracy, not talking about 1 cm?", "votes": 6, "replies": [ { "id": 659381, "postDate": "2019-10-27T14:54:17.310Z", "content": "I have same concern.\nRotation thresholds are 50 to 5 degree, relatively easy.
", "rawMarkdown": "I have same concern.\nRotation thresholds are 50 to 5 degree, relatively easy.", "votes": 2 } ] }, { "id": 656069, "postDate": "2019-10-23T21:40:35.470Z", "content": "Hi @philculliton @addisonhoward , I have another question : how do we map the \"model type\" encoded as a number in 'train.csv' to the corresponding JSON file car model?
", "rawMarkdown": "Hi @philculliton @addisonhoward , I have another question : how do we map the \"model type\" encoded as a number in 'train.csv' to the corresponding JSON file car model?", "votes": 6 }, { "id": 671862, "postDate": "2019-11-13T08:51:15.993Z", "content": "Is that possible to add more submission allowance please? We used to have 5.
", "rawMarkdown": "Is that possible to add more submission allowance please? We used to have 5.", "votes": 1 }, { "id": 655524, "postDate": "2019-10-23T06:11:26.737Z", "content": "Hi @philculliton , could you please tell me how to convert pickled files to obj files?So that I can open 3D models in meshlab.
", "rawMarkdown": "Hi @philculliton , could you please tell me how to convert pickled files to obj files?So that I can open 3D models in meshlab.", "votes": 1 }, { "id": 661912, "postDate": "2019-10-30T21:05:46.810Z", "content": "Hi, @philculliton as @ebouteillon mentioned previously we need a way to map JSON file to the \"model_type\" number.\nThe JSON files come in a different order for every new notebook.\nFor example in my case model_type 0 corresponds to ''jili-boyue.json'', model_type 1 to \"sikeda-jingrui.json\" .\nI suppose everyone got different values. \nThanks
", "rawMarkdown": "Hi, @philculliton as @ebouteillon mentioned previously we need a way to map JSON file to the \"model_type\" number.\nThe JSON files come in a different order for every new notebook.\nFor example in my case model_type 0 corresponds to ''jili-boyue.json'', model_type 1 to \"sikeda-jingrui.json\" .\nI suppose everyone got different values. \nThanks", "replies": [ { "id": 687047, "postDate": "2019-12-03T20:39:33.583Z", "content": "Maybe this is what you are looking for ? https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py
", "rawMarkdown": "Maybe this is what you are looking for ? [https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py](https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py)", "votes": 1 } ] }, { "id": 669367, "postDate": "2019-11-10T00:11:42.937Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 671013, "author_name": "Anton-br", "author_url": "", "post_date": "2019-11-12T06:34:52.310000", "content": "Hi @philculliton @addisonhoward! It is looks like score calculation is needed to be described more clearly because right now we can't understand how it works. I respectfully ask you to pay attention to this discussion (https://www.kaggle.com/c/pku-autonomous-driving/discussion/116661#latest-670626) and maybe provide us a code for metrics calculation.
\n\nFor example my first submission contains all predicted cars with confidence > 0.5 and score value were 0.079, but with confidence > 0.99 score is 0.169 (4th place of leaderboard right now) by the way this submission contains only 200 cars for all images.
", "votes": 11, "replies": [ { "id": 671650, "author_name": "stevenwudi", "author_url": "", "post_date": "2019-11-13T01:52:40.887000", "content": "200 cars for all images! Suppose 2000 test images has average of 10 cars per image, there will be around 20000 cars. Hence there is clearly an evaluation error for this competition. I would also suggest the competition host to provide the evaluation code.
", "votes": 3, "replies": [] } ] }, { "id": 655345, "author_name": "NathanO'Brien", "author_url": "", "post_date": "2019-10-23T00:37:46.573000", "content": "There seem to be a lot of cars missing from the mask or detection algorithm in the training model. Three examples;
\n\n| ID | Cars detected in train.csv | Comment |\n| --- | --- | --- |\n| ID_00ab59fa6 | 13 | There's a lot more than 13 cars in this photo, even excluding those masked |\n| ID_00ac30455 | 5 | There are 2 trucks in this photo, so I'm assuming they should just be ignored completely? |\n| ID_00b7fb303 | 4 | There is also a bus in this photo. Should that just be ignored? |
\n\nOnly submit for vehicles where there is a matching 3D model?\nWhat happens if you make a submission with more vehicles detected than your algorithm detects?
", "votes": 5, "replies": [] }, { "id": 655332, "author_name": "Eric Bouteillon", "author_url": "", "post_date": "2019-10-22T23:38:28.773000", "content": "Hi @philculliton , @addisonhoward , can you please indicate libraries we have to install to load pickled files? As the sample in description is not sufficient. Thanks in advance.
\n\n
@ebouteillon @philculliton I tried with different renders including pygame and PyOpenG modules and I couldn't even open the files. However, I think we'll not be able to load and visualize models on kaggle's kernels.\nI think the hosts should upload the proper code.
", "votes": 0, "replies": [] }, { "id": 655880, "author_name": "Phil Culliton", "author_url": "", "post_date": "2019-10-23T15:59:24.337000", "content": "Hi @ebouteillon @jesucristo - thanks! The pickled objects are just JSON blobs containing numpy arrays. I was able to un-pickle them on my own machines, but one of my coworkers also had a problem loading them.
\n\nIn lieu of trying to solve a bunch of pickle version issues, I've added the models as JSON files in the competition data. Let me know if that works all right, or if you still need help.
Hi @philculliton it is much better.
\n\nFor kagglers interested in seeing the models in 3D, I shared a notebook : https://www.kaggle.com/ebouteillon/load-a-3d-car-model
", "votes": 0, "replies": [] } ] }, { "id": 659344, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2019-10-27T13:42:53.360000", "content": "Thanks! Could you please double-check if distance thresholds on evaluation page are correct? I see \"Translation: 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01\" but that looks way too low, it's unlikely we'll have predictions with sub-10 cm accuracy, not talking about 1 cm?
", "votes": 6, "replies": [ { "id": 659381, "author_name": "s_shohei", "author_url": "", "post_date": "2019-10-27T14:54:17.310000", "content": "I have same concern.\nRotation thresholds are 50 to 5 degree, relatively easy.
", "votes": 2, "replies": [] } ] }, { "id": 656069, "author_name": "Eric Bouteillon", "author_url": "", "post_date": "2019-10-23T21:40:35.470000", "content": "Hi @philculliton @addisonhoward , I have another question : how do we map the \"model type\" encoded as a number in 'train.csv' to the corresponding JSON file car model?
", "votes": 6, "replies": [] }, { "id": 671862, "author_name": "zhibopotame", "author_url": "", "post_date": "2019-11-13T08:51:15.993000", "content": "Is that possible to add more submission allowance please? We used to have 5.
", "votes": 1, "replies": [] }, { "id": 655524, "author_name": "Linye Li", "author_url": "", "post_date": "2019-10-23T06:11:26.737000", "content": "Hi @philculliton , could you please tell me how to convert pickled files to obj files?So that I can open 3D models in meshlab.
", "votes": 1, "replies": [] }, { "id": 661912, "author_name": "Alex", "author_url": "", "post_date": "2019-10-30T21:05:46.810000", "content": "Hi, @philculliton as @ebouteillon mentioned previously we need a way to map JSON file to the \"model_type\" number.\nThe JSON files come in a different order for every new notebook.\nFor example in my case model_type 0 corresponds to ''jili-boyue.json'', model_type 1 to \"sikeda-jingrui.json\" .\nI suppose everyone got different values. \nThanks
", "votes": 0, "replies": [ { "id": 687047, "author_name": "Ilan Coulon", "author_url": "", "post_date": "2019-12-03T20:39:33.583000", "content": "Maybe this is what you are looking for ? https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py
", "votes": 1, "replies": [] } ] }, { "id": 669367, "author_name": "", "author_url": "", "post_date": "2019-11-10T00:11:42.937000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "671013": "Hi @philculliton @addisonhoward! It is looks like score calculation is needed to be described more clearly because right now we can't understand how it works. I respectfully ask you to pay attention to this discussion (https://www.kaggle.com/c/pku-autonomous-driving/discussion/116661#latest-670626) and maybe provide us a code for metrics calculation.\n\nFor example my first submission contains all predicted cars with confidence > 0.5 and score value were 0.079, but with confidence > 0.99 score is 0.169 (4th place of leaderboard right now) by the way this submission contains only 200 cars for all images.", "655214": "Welcome to the Peking University/Baidu - Autonomous Driving competition!\n\nIn this competition, we're predicting the positions and orientations of vehicles based on images taken by a single-camera test vehicle. Please check out the Evaluation and Data pages for information about how the metric works / the data is structured.\n\nThis should be fun / a little bit different - please let us know if you have questions!\n\nGood luck!", "655345": "There seem to be a lot of cars missing from the mask or detection algorithm in the training model. Three examples;\n\n| ID | Cars detected in train.csv | Comment |\n| --- | --- | --- |\n| ID_00ab59fa6 | 13 | There's a lot more than 13 cars in this photo, even excluding those masked |\n| ID_00ac30455 | 5 | There are 2 trucks in this photo, so I'm assuming they should just be ignored completely? |\n| ID_00b7fb303 | 4 | There is also a bus in this photo. Should that just be ignored? |\n\nOnly submit for vehicles where there is a matching 3D model?\nWhat happens if you make a submission with more vehicles detected than your algorithm detects?\n", "655332": "Hi @philculliton , @addisonhoward , can you please indicate libraries we have to install to load pickled files? As the sample in description is not sufficient. Thanks in advance.\n\n\n", "659344": "Thanks! Could you please double-check if distance thresholds on evaluation page are correct? I see \"Translation: 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01\" but that looks way too low, it's unlikely we'll have predictions with sub-10 cm accuracy, not talking about 1 cm?", "656069": "Hi @philculliton @addisonhoward , I have another question : how do we map the \"model type\" encoded as a number in 'train.csv' to the corresponding JSON file car model?", "671862": "Is that possible to add more submission allowance please? We used to have 5.", "655524": "Hi @philculliton , could you please tell me how to convert pickled files to obj files?So that I can open 3D models in meshlab.", "661912": "Hi, @philculliton as @ebouteillon mentioned previously we need a way to map JSON file to the \"model_type\" number.\nThe JSON files come in a different order for every new notebook.\nFor example in my case model_type 0 corresponds to ''jili-boyue.json'', model_type 1 to \"sikeda-jingrui.json\" .\nI suppose everyone got different values. \nThanks", "669367": "" } }