Model overfitting for 20 classes ( PASCAL VOC 2007 + 2012 dataset ) about machine-learning-collection HOT 3 CLOSED

I'm not sure what the problem could be, that sounds very strange. What kind of data augmentation are you using? I would try using quite extensive augmentation and probably use albumentations library. When I find some time I can take a look at this in more depth and see what performance I am able to get.

from machine-learning-collection.

I was using random blur , colorjitter . I even used extraction weights upon which yolo was trained. Didn't get better results . Now, i trained using resnet34 (pertained) . I got test map upto 40%.
May i will use albumentations library and train once and check .
And if you get some time . Please do let me know your results.

Thank you so much for your reply.😊

from machine-learning-collection.

I also got the problem that with this implementation I reach a pretty good mAP on the training data (VOC 2007 and 2012) after 135 epochs. But when I evaluate the model training on the test data I get a very (!) bad mAP... I read this multiple times (e.g. here: https://www.kaggle.com/code/vexxingbanana/yolov1-from-scratch-pytorch/comments) while I made some research to solve this topic. Unfortunately I wasn't able to do so yet... Maybe some of you guys got an idea?

from machine-learning-collection.