Reproduce results (DINO) on 10% label data on coco dataset

Open tahirashehzadi opened this issue 2 years ago • 4 comments

Hi, Thank you so much for your amazing work. I reproduce results on 10% label data on coco dataset for DINO network. However i am getting 37.2 mAP on 10% label while your result on paper shows 44.6 mAP on 10% label. Can you elaborate this.

Thanks in advance!

Apr 15 '24 11:04 tahirashehzadi

https://huggingface.co/czm369/MixPL/tree/main/mixpl_dino-4scale_r50_90k_coco-s1-p10.py

Apr 17 '24 04:04 Czm369

https://huggingface.co/czm369/MixPL/tree/main/mixpl_dino-4scale_r50_90k_coco-s1-p10.py

Hi,

Your results are amazing! I would appreciate if you could update README.MD file. I used the same file but I am not able to run the DINO model. Also, for the Faster-RCNN, I could not reproduce the results even with different learning rates and batch sizes. Could you please share some more insights like what dataset you used, did you use parallel distributed training?

Thanks, Bharani.

Jul 25 '24 13:07 bharanibala