Ruijie Yao
Ruijie Yao
Me too, I meet the same problem. Maybe we need to use torch.Sigmoid(outputs).
We also need a Imagenet-21K pretrained model. Do you have a plan to provide it? @iamhankai
I also had a hard time and mine is 83.1 using resnet101 ,and I followed the tricks in https://github.com/Alibaba-MIIL/ASL/issues/30#issuecomment-750780576
The file of map=83.1 was deleted these, so I ran the experiment again with batch_size=6 per GPU and 48 1080Ti GPUS. The new result is mAP=82.51. I use MMclassification and...
The training log is: -------------------- 2022-07-13 10:38:26,914 - mmcls - INFO - workflow: [('train', 1)], max: 80 epochs 2022-07-13 10:38:26,978 - mmcls - INFO - Checkpoints will be saved to...
Sure, I have updated my email address.
Me too. I strongly agree!
@ThiagoViek Remarkable ideas! Have you tried some experiments? Expecting for some feedback.