Hi all...\nI have done projects using deep learning algorithms like CNN and RNN. Can someone point out where to get started for solving this challenge.
", "messages": [ { "id": "677804", "postDate": "11/20/2019 15:51:38", "content": "Hi all...\nI have done projects using deep learning algorithms like CNN and RNN. Can someone point out where to get started for solving this challenge.
", "rawMarkdown": "Hi all...\nI have done projects using deep learning algorithms like CNN and RNN. Can someone point out where to get started for solving this challenge.", "votes": null }, { "id": "677866", "postDate": "11/20/2019 17:20:50", "content": "This is a different beast altogether . I would suggest first read a bit about instance segmentation , mask rcnn , bounding boxes in 2D etc . After that look into 3D segmentation . Apolloscape papers , Kitti dataset and papers etc .
", "rawMarkdown": "This is a different beast altogether . I would suggest first read a bit about instance segmentation , mask rcnn , bounding boxes in 2D etc . After that look into 3D segmentation . Apolloscape papers , Kitti dataset and papers etc .", "votes": null }, { "id": "677871", "postDate": "11/20/2019 17:29:50", "content": "I don't think any segmentation is needed for this competition, the submission file only shows box coordinates and the classification of the car model.
\n\nFrom what I understand the objective is to fit the boxes and classify the car model. We don't need to predict segmentation masks.
\n\nIn the data description it says:
\n\n\n", "rawMarkdown": "I don't think any segmentation is needed for this competition, the submission file only shows box coordinates and the classification of the car model.\n\n From what I understand the objective is to fit the boxes and classify the car model. We don't need to predict segmentation masks. \n\nIn the data description it says:\n> train_masks.zip - ignore masks for the training set. (Not all images have a mask.)\n> test_masks.zip - ignore masks for the test set. (Not all images have a mask.)", "votes": null }, { "id": "677874", "postDate": "11/20/2019 17:35:27", "content": "train_masks.zip - ignore masks for the training set. (Not all images have a mask.)\n test_masks.zip - ignore masks for the test set. (Not all images have a mask.)
\n
No Jack , we don't have to predict the car model I guess ,but we need to estimate the pose of the car . I think a little bit idea regarding segmentation would be better for a beginner because before doing a couple of segmentation competition I didn't even know what is mask and what is bounding box .
", "rawMarkdown": "No Jack , we don't have to predict the car model I guess ,but we need to estimate the pose of the car . I think a little bit idea regarding segmentation would be better for a beginner because before doing a couple of segmentation competition I didn't even know what is mask and what is bounding box .", "votes": null }, { "id": "677880", "postDate": "11/20/2019 17:48:45", "content": "Thanks..👍
", "rawMarkdown": "Thanks..👍", "votes": null }, { "id": "677882", "postDate": "11/20/2019 17:49:55", "content": "My mistake, but I am happy to hear we don't need to predict the car model. That would be very difficult. You are correct, it definitely helps to know the difference between segmentation, bounding box etc..
\n\nThis courses is really good and covers all those topics https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLfKuNRZBWLruVvONn3LwC6m21JSAGUGuU
", "rawMarkdown": "My mistake, but I am happy to hear we don't need to predict the car model. That would be very difficult. You are correct, it definitely helps to know the difference between segmentation, bounding box etc..\n\nThis courses is really good and covers all those topics https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLfKuNRZBWLruVvONn3LwC6m21JSAGUGuU", "votes": null }, { "id": "679105", "postDate": "11/22/2019 09:40:26", "content": "Some simple ideas: add pose head to predict yaw pitch roll on Faster R-CNN
", "rawMarkdown": "Some simple ideas: add pose head to predict yaw pitch roll on Faster R-CNN", "votes": null }, { "id": "679144", "postDate": "11/22/2019 10:17:13", "content": "Exactly .
", "rawMarkdown": "Exactly .", "votes": null }, { "id": "696530", "postDate": "12/16/2019 18:19:19", "content": "@zstusnoopy Hi! Are you intrested in teaming up? I have a 2080Ti and completed competitions previously with decent scores. I also have a pretty good LB score now. I think we can build good solutions together. Thanks!
", "rawMarkdown": "zstusnoopy Hi! Are you intrested in teaming up? I have a 2080Ti and completed competitions previously with decent scores. I also have a pretty good LB score now. I think we can build good solutions together. Thanks!", "votes": null }, { "id": "696820", "postDate": "12/17/2019 05:29:01", "content": "Hi！Thank you for your invite, but I have no plans to form a team at present.
", "rawMarkdown": "Hi！Thank you for your invite, but I have no plans to form a team at present.", "votes": null }, { "id": "696866", "postDate": "12/17/2019 06:59:41", "content": "It's fine! Thank you!
", "rawMarkdown": "It's fine! Thank you!", "votes": null }, { "id": "717699", "postDate": "01/13/2020 13:44:10", "content": "@zstusnoopy So, i need to edit Faster R-CNN architecture to predict yaw, roll pitch or make use of predictions from Faster R-CNN to regress other values ?
", "rawMarkdown": "zstusnoopy So, i need to edit Faster R-CNN architecture to predict yaw, roll pitch or make use of predictions from Faster R-CNN to regress other values ?", "votes": null }, { "id": "718179", "postDate": "01/14/2020 06:06:52", "content": "yes
", "rawMarkdown": "yes", "votes": null } ], "comments": [ { "id": 677866, "author_name": "phoenix9032", "author_url": "", "post_date": "11/20/2019 17:20:50", "content": "This is a different beast altogether . I would suggest first read a bit about instance segmentation , mask rcnn , bounding boxes in 2D etc . After that look into 3D segmentation . Apolloscape papers , Kitti dataset and papers etc .
", "votes": null, "replies": [ { "id": 677871, "author_name": "jackvial", "author_url": "", "post_date": "11/20/2019 17:29:50", "content": "I don't think any segmentation is needed for this competition, the submission file only shows box coordinates and the classification of the car model.
\n\nFrom what I understand the objective is to fit the boxes and classify the car model. We don't need to predict segmentation masks.
\n\nIn the data description it says:
\n\n\n", "votes": null, "replies": [] }, { "id": 677874, "author_name": "phoenix9032", "author_url": "", "post_date": "11/20/2019 17:35:27", "content": "train_masks.zip - ignore masks for the training set. (Not all images have a mask.)\n test_masks.zip - ignore masks for the test set. (Not all images have a mask.)
\n
No Jack , we don't have to predict the car model I guess ,but we need to estimate the pose of the car . I think a little bit idea regarding segmentation would be better for a beginner because before doing a couple of segmentation competition I didn't even know what is mask and what is bounding box .
", "votes": null, "replies": [] }, { "id": 677880, "author_name": "prayagmadhu", "author_url": "", "post_date": "11/20/2019 17:48:45", "content": "Thanks..👍
", "votes": null, "replies": [] }, { "id": 677882, "author_name": "jackvial", "author_url": "", "post_date": "11/20/2019 17:49:55", "content": "My mistake, but I am happy to hear we don't need to predict the car model. That would be very difficult. You are correct, it definitely helps to know the difference between segmentation, bounding box etc..
\n\nThis courses is really good and covers all those topics https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLfKuNRZBWLruVvONn3LwC6m21JSAGUGuU
", "votes": null, "replies": [] } ] }, { "id": 679105, "author_name": "zstusnoopy", "author_url": "", "post_date": "11/22/2019 09:40:26", "content": "Some simple ideas: add pose head to predict yaw pitch roll on Faster R-CNN
", "votes": null, "replies": [ { "id": 679144, "author_name": "phoenix9032", "author_url": "", "post_date": "11/22/2019 10:17:13", "content": "Exactly .
", "votes": null, "replies": [] }, { "id": 696530, "author_name": "tonychenxyz", "author_url": "", "post_date": "12/16/2019 18:19:19", "content": "@zstusnoopy Hi! Are you intrested in teaming up? I have a 2080Ti and completed competitions previously with decent scores. I also have a pretty good LB score now. I think we can build good solutions together. Thanks!
", "votes": null, "replies": [] }, { "id": 696820, "author_name": "zstusnoopy", "author_url": "", "post_date": "12/17/2019 05:29:01", "content": "Hi！Thank you for your invite, but I have no plans to form a team at present.
", "votes": null, "replies": [] }, { "id": 696866, "author_name": "tonychenxyz", "author_url": "", "post_date": "12/17/2019 06:59:41", "content": "It's fine! Thank you!
", "votes": null, "replies": [] } ] }, { "id": 717699, "author_name": "prayagmadhu", "author_url": "", "post_date": "01/13/2020 13:44:10", "content": "@zstusnoopy So, i need to edit Faster R-CNN architecture to predict yaw, roll pitch or make use of predictions from Faster R-CNN to regress other values ?
", "votes": null, "replies": [ { "id": 718179, "author_name": "zstusnoopy", "author_url": "", "post_date": "01/14/2020 06:06:52", "content": "yes
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "677804": "Hi all...\nI have done projects using deep learning algorithms like CNN and RNN. Can someone point out where to get started for solving this challenge.", "677866": "This is a different beast altogether . I would suggest first read a bit about instance segmentation , mask rcnn , bounding boxes in 2D etc . After that look into 3D segmentation . Apolloscape papers , Kitti dataset and papers etc .", "677871": "I don't think any segmentation is needed for this competition, the submission file only shows box coordinates and the classification of the car model.\n\n From what I understand the objective is to fit the boxes and classify the car model. We don't need to predict segmentation masks. \n\nIn the data description it says:\n> train_masks.zip - ignore masks for the training set. (Not all images have a mask.)\n> test_masks.zip - ignore masks for the test set. (Not all images have a mask.)", "677874": "No Jack , we don't have to predict the car model I guess ,but we need to estimate the pose of the car . I think a little bit idea regarding segmentation would be better for a beginner because before doing a couple of segmentation competition I didn't even know what is mask and what is bounding box .", "677880": "Thanks..👍", "677882": "My mistake, but I am happy to hear we don't need to predict the car model. That would be very difficult. You are correct, it definitely helps to know the difference between segmentation, bounding box etc..\n\nThis courses is really good and covers all those topics https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLfKuNRZBWLruVvONn3LwC6m21JSAGUGuU", "679105": "Some simple ideas: add pose head to predict yaw pitch roll on Faster R-CNN", "679144": "Exactly .", "696530": "zstusnoopy Hi! Are you intrested in teaming up? I have a 2080Ti and completed competitions previously with decent scores. I also have a pretty good LB score now. I think we can build good solutions together. Thanks!", "696820": "Hi！Thank you for your invite, but I have no plans to form a team at present.", "696866": "It's fine! Thank you!", "717699": "zstusnoopy So, i need to edit Faster R-CNN architecture to predict yaw, roll pitch or make use of predictions from Faster R-CNN to regress other values ?", "718179": "yes" }, "source": "meta" }