Now you can train and test models using YoloR. Both notebooks are prepared for this competition - you can create modes on COTS dataset and submit score to LB. As I said these notebooks are for development - are not solutions (this is why score is not high). In my opinion these notebooks are great shortcut for anyone who wants to start or get to know YoloR. Only way to progress is experiment.
\nTraining: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-train
\nInference + competition submission: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-infer
enjoy
\nThis is our (as a team) next contribution to this competition. Previous one:
\n", "messages": [ { "id": "1639569", "postDate": "01/05/2022 18:49:46", "content": "Now you can train and test models using YoloR. Both notebooks are prepared for this competition - you can create modes on COTS dataset and submit score to LB. As I said these notebooks are for development - are not solutions (this is why score is not high). In my opinion these notebooks are great shortcut for anyone who wants to start or get to know YoloR. Only way to progress is experiment.
\nTraining: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-train
\nInference + competition submission: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-infer
enjoy
\nThis is our (as a team) next contribution to this competition. Previous one:
\n", "rawMarkdown": "Now you can train and test models using YoloR. Both notebooks are prepared for this competition - you can create modes on COTS dataset and submit score to LB. As I said these notebooks are for development - are not solutions (this is why score is not high). In my opinion these notebooks are great shortcut for anyone who wants to start or get to know YoloR. Only way to progress is experiment.\n\nTraining: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-train\nInference + competition submission: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-infer\n\nenjoy \n\nThis is our (as a team) next contribution to this competition. Previous one:\n- [YoloX full training pipeline for COTS dataset](https://www.kaggle.com/remekkinas/yolox-training-pipeline-cots-dataset-lb-0-507?scriptVersionId=81353936)\n- [YoloX detections submission made on COTS dataset](https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots-lb-0-507)", "votes": null }, { "id": "1639614", "postDate": "01/05/2022 19:32:36", "content": "@imuhammadismail Thank you very much! The most tricky part was inference. Most examples (all) use standard detect.py script proived by YoloR. Now you can experiment as you can :) and subimit using Kaggle API.
\nAs you see … you can share and be on TOP. I know that it will be change (we have many great Kagglers in this competition) but I am trying to contribute. I know that many Kaggles need some help and notebooks to start. This is why I created this set.
\nWe will see what model finally win … YoloX, Yolov5, YoloR … and maybe else. I do not know … but I want to give people possibility to experiment without any limitations. We have Yolo5 notebooks provided by @awsaf49 and two mine YoloX and YoloR.
", "rawMarkdown": "imuhammadismail Thank you very much! The most tricky part was inference. Most examples (all) use standard detect.py script proived by YoloR. Now you can experiment as you can :) and subimit using Kaggle API.\n\nAs you see ... you can share and be on TOP. I know that it will be change (we have many great Kagglers in this competition) but I am trying to contribute. I know that many Kaggles need some help and notebooks to start. This is why I created this set.\n\nWe will see what model finally win ... YoloX, Yolov5, YoloR ... and maybe else. I do not know ... but I want to give people possibility to experiment without any limitations. We have Yolo5 notebooks provided by @awsaf49 and two mine YoloX and YoloR.", "votes": null }, { "id": "1639932", "postDate": "01/06/2022 04:23:28", "content": "@remekkinas many thanks to you for sharing your work and learning with all.
\nIf you get a chance please also try and share the train/submission using [https://github.com/AlexeyAB/darknet/issues/7087](Darknet's scaled YOLO).
I am struggling with the submission part.
\nI have made my train and submission notebook public so that in case you get a chance to have a look at my submission part-https://www.kaggle.com/iamprateek/cots-yolov4-train/notebook and https://www.kaggle.com/iamprateek/cots-yolov4-submission/notebook
thx for sharing.
\ncollab pro needs keeping interactive, that bothers. currently no budget for colab pro+. Kaggle gpu hrs is limited.
", "rawMarkdown": "thx for sharing. \n\ncollab pro needs keeping interactive, that bothers. currently no budget for colab pro+. Kaggle gpu hrs is limited.", "votes": null }, { "id": "1640384", "postDate": "01/06/2022 12:50:41", "content": "But now you can use it … :)
\nThank you for vote :)
I will look into it soon. I think that there is no problem to run yolo scaled on Kaggle. Let me some time …
", "rawMarkdown": "I will look into it soon. I think that there is no problem to run yolo scaled on Kaggle. Let me some time ...", "votes": null }, { "id": "1640618", "postDate": "01/06/2022 16:29:45", "content": "Thank you for your kind sharing.
\nIt costs too much time to try bigger model or tune the parameters without cutting edge hardware.
", "rawMarkdown": "Thank you for your kind sharing.\n\nIt costs too much time to try bigger model or tune the parameters without cutting edge hardware.", "votes": null }, { "id": "1641156", "postDate": "01/07/2022 06:37:07", "content": "Thanks for sharing amazing notebooks and discussions. I have a question about dataset. Random split or kfold sequence split which better make sense for this task.
", "rawMarkdown": "Thanks for sharing amazing notebooks and discussions. I have a question about dataset. Random split or kfold sequence split which better make sense for this task.", "votes": null }, { "id": "1641218", "postDate": "01/07/2022 07:55:00", "content": "This is important question. Answer is simple - try. Why I am answering such way? Because this is only way you can learn and enjoy learning and discovery process. Trust me … :)
", "rawMarkdown": "This is important question. Answer is simple - try. Why I am answering such way? Because this is only way you can learn and enjoy learning and discovery process. Trust me … :)", "votes": null } ], "comments": [ { "id": 1639614, "author_name": "imuhammadismail", "author_url": "", "post_date": "01/05/2022 19:32:36", "content": "@imuhammadismail Thank you very much! The most tricky part was inference. Most examples (all) use standard detect.py script proived by YoloR. Now you can experiment as you can :) and subimit using Kaggle API.
\nAs you see … you can share and be on TOP. I know that it will be change (we have many great Kagglers in this competition) but I am trying to contribute. I know that many Kaggles need some help and notebooks to start. This is why I created this set.
\nWe will see what model finally win … YoloX, Yolov5, YoloR … and maybe else. I do not know … but I want to give people possibility to experiment without any limitations. We have Yolo5 notebooks provided by @awsaf49 and two mine YoloX and YoloR.
", "votes": null, "replies": [] }, { "id": 1640091, "author_name": "dragonzhang", "author_url": "", "post_date": "01/06/2022 07:01:54", "content": "thx for sharing.
\ncollab pro needs keeping interactive, that bothers. currently no budget for colab pro+. Kaggle gpu hrs is limited.
", "votes": null, "replies": [] }, { "id": 1640384, "author_name": "remekkinas", "author_url": "", "post_date": "01/06/2022 12:50:41", "content": "But now you can use it … :)
\nThank you for vote :)
Thank you for your kind sharing.
\nIt costs too much time to try bigger model or tune the parameters without cutting edge hardware.
", "votes": null, "replies": [] } ] }, { "id": 1639932, "author_name": "iamprateek", "author_url": "", "post_date": "01/06/2022 04:23:28", "content": "@remekkinas many thanks to you for sharing your work and learning with all.
\nIf you get a chance please also try and share the train/submission using [https://github.com/AlexeyAB/darknet/issues/7087](Darknet's scaled YOLO).
I am struggling with the submission part.
\nI have made my train and submission notebook public so that in case you get a chance to have a look at my submission part-https://www.kaggle.com/iamprateek/cots-yolov4-train/notebook and https://www.kaggle.com/iamprateek/cots-yolov4-submission/notebook
I will look into it soon. I think that there is no problem to run yolo scaled on Kaggle. Let me some time …
", "votes": null, "replies": [] } ] }, { "id": 1641156, "author_name": "ichimarugin", "author_url": "", "post_date": "01/07/2022 06:37:07", "content": "Thanks for sharing amazing notebooks and discussions. I have a question about dataset. Random split or kfold sequence split which better make sense for this task.
", "votes": null, "replies": [ { "id": 1641218, "author_name": "remekkinas", "author_url": "", "post_date": "01/07/2022 07:55:00", "content": "This is important question. Answer is simple - try. Why I am answering such way? Because this is only way you can learn and enjoy learning and discovery process. Trust me … :)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1639569": "Now you can train and test models using YoloR. Both notebooks are prepared for this competition - you can create modes on COTS dataset and submit score to LB. As I said these notebooks are for development - are not solutions (this is why score is not high). In my opinion these notebooks are great shortcut for anyone who wants to start or get to know YoloR. Only way to progress is experiment.\n\nTraining: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-train\nInference + competition submission: https://www.kaggle.com/remekkinas/yolor-p6-w6-one-more-yolo-on-kaggle-infer\n\nenjoy \n\nThis is our (as a team) next contribution to this competition. Previous one:\n- [YoloX full training pipeline for COTS dataset](https://www.kaggle.com/remekkinas/yolox-training-pipeline-cots-dataset-lb-0-507?scriptVersionId=81353936)\n- [YoloX detections submission made on COTS dataset](https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots-lb-0-507)", "1639614": "### Hi [Remek Kinas](remekkinas), Thank for sharing your impressive work on YoloR. I hope other kagglers get maximum benefits from your work.\n### Also I recommend this video on [How to Train YOLOR on a Custom Dataset](https://www.youtube.com/watch?v=sZ5DiXDOHEM) to all those kaggler who are interested in YOLOR", "1639623": "imuhammadismail Thank you very much! The most tricky part was inference. Most examples (all) use standard detect.py script proived by YoloR. Now you can experiment as you can :) and subimit using Kaggle API.\n\nAs you see ... you can share and be on TOP. I know that it will be change (we have many great Kagglers in this competition) but I am trying to contribute. I know that many Kaggles need some help and notebooks to start. This is why I created this set.\n\nWe will see what model finally win ... YoloX, Yolov5, YoloR ... and maybe else. I do not know ... but I want to give people possibility to experiment without any limitations. We have Yolo5 notebooks provided by @awsaf49 and two mine YoloX and YoloR.", "1639932": "remekkinas many thanks to you for sharing your work and learning with all. \nIf you get a chance please also try and share the train/submission using [https://github.com/AlexeyAB/darknet/issues/7087](Darknet's scaled YOLO). \n\nI am struggling with the submission part.\nI have made my train and submission notebook public so that in case you get a chance to have a look at my submission part-[https://www.kaggle.com/iamprateek/cots-yolov4-train/notebook](Train) and [https://www.kaggle.com/iamprateek/cots-yolov4-submission/notebook](Submission)", "1640091": "thx for sharing. \n\ncollab pro needs keeping interactive, that bothers. currently no budget for colab pro+. Kaggle gpu hrs is limited.", "1640384": "But now you can use it ... :) \nThank you for vote :)", "1640385": "I will look into it soon. I think that there is no problem to run yolo scaled on Kaggle. Let me some time ...", "1640618": "Thank you for your kind sharing.\n\nIt costs too much time to try bigger model or tune the parameters without cutting edge hardware.", "1641156": "Thanks for sharing amazing notebooks and discussions. I have a question about dataset. Random split or kfold sequence split which better make sense for this task.", "1641218": "This is important question. Answer is simple - try. Why I am answering such way? Because this is only way you can learn and enjoy learning and discovery process. Trust me … :)" }, "source": "meta" }