![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F998389%2F2fef8fb10bc00c3ec50631eaf060a839%2F_20230623145903.png?generation=1687500231667261&alt=media\")
I trained my yolo v7 model based shared notebook. But the logs with W&B shows that it is a little bit strange.
\nIs there possible reason that can cause this?
I trained my yolo v7 model based shared notebook. But the logs with W&B shows that it is a little bit strange.
\nIs there possible reason that can cause this?
I spotted it as well.
\nIt looks that the decrease in precision is accompanied by an increase in recall, so at some point the network is deciding it would rather catch more of the true positives at the cost of more false positives, which will make recall increase and precision drop.
If this is a variant of my notebook, are you then training with all 3 classes?
\nI'm wondering if it wouldn't make sense to only use the blood vessel class.
\nCould be that the network at some point decides to classify some of the \"unsure\" class as blood vessels, leading to a drop in precision.
\nBut just hypothesising.
Thank you fnands. I am using your posted notebook. Thanks again for your kind sharing and reply.
\nI will try to use a model trained with only class1 and another model trained with only class 2 to check if it would help.
", "rawMarkdown": "Thank you fnands. I am using your posted notebook. Thanks again for your kind sharing and reply.\n\nI will try to use a model trained with only class1 and another model trained with only class 2 to check if it would help.", "votes": null } ], "comments": [ { "id": 2314149, "author_name": "fnands", "author_url": "", "post_date": "06/23/2023 07:57:31", "content": "I spotted it as well.
\nIt looks that the decrease in precision is accompanied by an increase in recall, so at some point the network is deciding it would rather catch more of the true positives at the cost of more false positives, which will make recall increase and precision drop.
If this is a variant of my notebook, are you then training with all 3 classes?
\nI'm wondering if it wouldn't make sense to only use the blood vessel class.
\nCould be that the network at some point decides to classify some of the \"unsure\" class as blood vessels, leading to a drop in precision.
\nBut just hypothesising.
Thank you fnands. I am using your posted notebook. Thanks again for your kind sharing and reply.
\nI will try to use a model trained with only class1 and another model trained with only class 2 to check if it would help.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2314046": "\n\nI trained my yolo v7 model based shared notebook. But the logs with W&B shows that it is a little bit strange. \nIs there possible reason that can cause this?\n\nhttps://wandb.ai/kw/yolov7-fine-tune/reports/metrics-precision-B-23-06-23-14-56-15---Vmlldzo0NzE4NzE2", "2314149": "I spotted it as well.\nIt looks that the decrease in precision is accompanied by an increase in recall, so at some point the network is deciding it would rather catch more of the true positives at the cost of more false positives, which will make recall increase and precision drop. \n\nIf this is a variant of my notebook, are you then training with all 3 classes? \nI'm wondering if it wouldn't make sense to only use the blood vessel class. \nCould be that the network at some point decides to classify some of the \"unsure\" class as blood vessels, leading to a drop in precision. \nBut just hypothesising.", "2314162": "Thank you fnands. I am using your posted notebook. Thanks again for your kind sharing and reply.\n\nI will try to use a model trained with only class1 and another model trained with only class 2 to check if it would help." }, "source": "meta" }