Zigarss
Zigarss
Same issue as them,I uesd the 8.0.22 version and 8.0.23 version, but both of them doesn't work. the same error occured in docker images and CLI. ```bash yolo task=detect mode=train...
期待支持segment！！
Hi, @auvipy @jan-zmeskal I have this problem too. my celery version is 5.3.0, and used redis as the broker and backend. task state was tend to "PENDING" in **24h**. although...
@gathierry Hi, I reconstruct your fastflow code, and my wide_resnet50_2 have the 41.33M(paper:41.3M) A.D. Param, and resnet18 have 4.65M(paper:4.9M) A.D. Param. Also the cait and deit have the same A.D....
@gathierry I just seperate the model to encoder(feature_extractor) and decoder(fastflow A.D.) like c-flow, just calculate the decoder's A.D. Param in model loading, and I get the right 41.3M in wrn50...
@gathierry Yes, I put the LayerNorm in the encoder, maybe the original paper also did this. because without LayerNorm, the decoder(FastFlow) can match the paper's A.D. Params. After all, the...