I am using yolov5 for this project, but I am trying to figure out what model to use for training. However, I'm trying to decide what to use as a model to train on, because yolov5s is not very accurate, and yolov5x takes too much time and uses a lot of memory. It would be great if you could tell me what is the best balanced solution at the moment. By the way, I'm using Google colab+ for training.
", "messages": [ { "id": "1656134", "postDate": "01/19/2022 05:36:22", "content": "I am using yolov5 for this project, but I am trying to figure out what model to use for training. However, I'm trying to decide what to use as a model to train on, because yolov5s is not very accurate, and yolov5x takes too much time and uses a lot of memory. It would be great if you could tell me what is the best balanced solution at the moment. By the way, I'm using Google colab+ for training.
", "rawMarkdown": "I am using yolov5 for this project, but I am trying to figure out what model to use for training. However, I'm trying to decide what to use as a model to train on, because yolov5s is not very accurate, and yolov5x takes too much time and uses a lot of memory. It would be great if you could tell me what is the best balanced solution at the moment. By the way, I'm using Google colab+ for training.", "votes": null }, { "id": "1656161", "postDate": "01/19/2022 06:04:27", "content": "I using yolov5m, i think it's good enough to ~0.6 lb.
", "rawMarkdown": "I using yolov5m, i think it's good enough to ~0.6 lb.", "votes": null }, { "id": "1656246", "postDate": "01/19/2022 07:52:54", "content": "Thanks for the reply. I'm going to give it another try with yolov5m. I think the balance between computational cost and model accuracy is both difficult and interesting in this field. As long as I have the computational cost and the time until the deadline, I will not stop trying. By the way, are you planning to challenge with yolo5vl or yolo5vx?
", "rawMarkdown": "Thanks for the reply. I'm going to give it another try with yolov5m. I think the balance between computational cost and model accuracy is both difficult and interesting in this field. As long as I have the computational cost and the time until the deadline, I will not stop trying. By the way, are you planning to challenge with yolo5vl or yolo5vx?", "votes": null }, { "id": "1656603", "postDate": "01/19/2022 13:24:49", "content": "Could share some advice?I am also use yolov5m.But the best score is 0.567.
", "rawMarkdown": "Could share some advice?I am also use yolov5m.But the best score is 0.567.", "votes": null }, { "id": "1656863", "postDate": "01/19/2022 17:47:55", "content": "How about yolov5m6, had you tried it?
", "rawMarkdown": "How about yolov5m6, had you tried it?", "votes": null }, { "id": "1657585", "postDate": "01/20/2022 10:01:52", "content": "In my case, i try experiments on small model, test best suited hypeparamters and then try it on larger model. This is my first time on kaggle competition so i don't know much, but as a learner i only have an advice is run as much as possible experiments. 👍 (you should try other techniques like cross folding, ensembles, or tracking). have fun learning.
", "rawMarkdown": "In my case, i try experiments on small model, test best suited hypeparamters and then try it on larger model. This is my first time on kaggle competition so i don't know much, but as a learner i only have an advice is run as much as possible experiments. 👍 (you should try other techniques like cross folding, ensembles, or tracking). have fun learning.", "votes": null }, { "id": "1658445", "postDate": "01/21/2022 03:14:06", "content": "There are many guys use yolo5s6, and also can get a good lb
", "rawMarkdown": "There are many guys use yolo5s6, and also can get a good lb", "votes": null }, { "id": "1658460", "postDate": "01/21/2022 03:40:30", "content": "Do you have any suggestions on how to find the best hyperparameters? I tried evolution but it didn’t help.
", "rawMarkdown": "Do you have any suggestions on how to find the best hyperparameters? I tried evolution but it didn’t help.", "votes": null }, { "id": "1659661", "postDate": "01/22/2022 03:16:17", "content": "Oh, I see. I'd like to know how they improve the accuracy.😳
", "rawMarkdown": "Oh, I see. I'd like to know how they improve the accuracy.😳", "votes": null }, { "id": "1660028", "postDate": "01/22/2022 10:28:11", "content": "Very curious about how to use yolov5m6 to gain 0.6 LB, such amazing!
", "rawMarkdown": "Very curious about how to use yolov5m6 to gain 0.6 LB, such amazing!", "votes": null }, { "id": "1660042", "postDate": "01/22/2022 10:49:32", "content": "Agree, I'm still using yolov5s6😅, m6 performed better locally, but quite bad on the LB
", "rawMarkdown": "Agree, I'm still using yolov5s6😅, m6 performed better locally, but quite bad on the LB", "votes": null }, { "id": "1680784", "postDate": "02/08/2022 03:03:05", "content": "Are anchor boxes one of your tuned hyperparameters or you rely more on k-mean clustering readily available by ultralytics?
", "rawMarkdown": "Are anchor boxes one of your tuned hyperparameters or you rely more on k-mean clustering readily available by ultralytics?", "votes": null } ], "comments": [ { "id": 1656161, "author_name": "locbaop", "author_url": "", "post_date": "01/19/2022 06:04:27", "content": "I using yolov5m, i think it's good enough to ~0.6 lb.
", "votes": null, "replies": [ { "id": 1656246, "author_name": "cahein1371", "author_url": "", "post_date": "01/19/2022 07:52:54", "content": "Thanks for the reply. I'm going to give it another try with yolov5m. I think the balance between computational cost and model accuracy is both difficult and interesting in this field. As long as I have the computational cost and the time until the deadline, I will not stop trying. By the way, are you planning to challenge with yolo5vl or yolo5vx?
", "votes": null, "replies": [] }, { "id": 1656603, "author_name": "rishengyang", "author_url": "", "post_date": "01/19/2022 13:24:49", "content": "Could share some advice?I am also use yolov5m.But the best score is 0.567.
", "votes": null, "replies": [] }, { "id": 1656863, "author_name": "w3579628328", "author_url": "", "post_date": "01/19/2022 17:47:55", "content": "How about yolov5m6, had you tried it?
", "votes": null, "replies": [] }, { "id": 1657585, "author_name": "locbaop", "author_url": "", "post_date": "01/20/2022 10:01:52", "content": "In my case, i try experiments on small model, test best suited hypeparamters and then try it on larger model. This is my first time on kaggle competition so i don't know much, but as a learner i only have an advice is run as much as possible experiments. 👍 (you should try other techniques like cross folding, ensembles, or tracking). have fun learning.
", "votes": null, "replies": [] }, { "id": 1658460, "author_name": "ayu055", "author_url": "", "post_date": "01/21/2022 03:40:30", "content": "Do you have any suggestions on how to find the best hyperparameters? I tried evolution but it didn’t help.
", "votes": null, "replies": [] }, { "id": 1660028, "author_name": "calvchen", "author_url": "", "post_date": "01/22/2022 10:28:11", "content": "Very curious about how to use yolov5m6 to gain 0.6 LB, such amazing!
", "votes": null, "replies": [] }, { "id": 1660042, "author_name": "w3579628328", "author_url": "", "post_date": "01/22/2022 10:49:32", "content": "Agree, I'm still using yolov5s6😅, m6 performed better locally, but quite bad on the LB
", "votes": null, "replies": [] }, { "id": 1680784, "author_name": "anhnhunhat", "author_url": "", "post_date": "02/08/2022 03:03:05", "content": "Are anchor boxes one of your tuned hyperparameters or you rely more on k-mean clustering readily available by ultralytics?
", "votes": null, "replies": [] } ] }, { "id": 1658445, "author_name": "ezkingt0t", "author_url": "", "post_date": "01/21/2022 03:14:06", "content": "There are many guys use yolo5s6, and also can get a good lb
", "votes": null, "replies": [ { "id": 1659661, "author_name": "cahein1371", "author_url": "", "post_date": "01/22/2022 03:16:17", "content": "Oh, I see. I'd like to know how they improve the accuracy.😳
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1656134": "I am using yolov5 for this project, but I am trying to figure out what model to use for training. However, I'm trying to decide what to use as a model to train on, because yolov5s is not very accurate, and yolov5x takes too much time and uses a lot of memory. It would be great if you could tell me what is the best balanced solution at the moment. By the way, I'm using Google colab+ for training.", "1656161": "I using yolov5m, i think it's good enough to ~0.6 lb.", "1656246": "Thanks for the reply. I'm going to give it another try with yolov5m. I think the balance between computational cost and model accuracy is both difficult and interesting in this field. As long as I have the computational cost and the time until the deadline, I will not stop trying. By the way, are you planning to challenge with yolo5vl or yolo5vx?", "1656603": "Could share some advice?I am also use yolov5m.But the best score is 0.567.", "1656863": "How about yolov5m6, had you tried it?", "1657585": "In my case, i try experiments on small model, test best suited hypeparamters and then try it on larger model. This is my first time on kaggle competition so i don't know much, but as a learner i only have an advice is run as much as possible experiments. 👍 (you should try other techniques like cross folding, ensembles, or tracking). have fun learning.", "1658445": "There are many guys use yolo5s6, and also can get a good lb", "1658460": "Do you have any suggestions on how to find the best hyperparameters? I tried evolution but it didn’t help.", "1659661": "Oh, I see. I'd like to know how they improve the accuracy.😳", "1660028": "Very curious about how to use yolov5m6 to gain 0.6 LB, such amazing!", "1660042": "Agree, I'm still using yolov5s6😅, m6 performed better locally, but quite bad on the LB", "1680784": "Are anchor boxes one of your tuned hyperparameters or you rely more on k-mean clustering readily available by ultralytics?" }, "source": "meta" }