Zhe Chen

Hi, thanks very much for your code. May I ask if you have met this problem?

> > Hi, thanks very much for your code. May I ask if you have met this problem? > > > > When did this happen ? at the beginning...

@Li-Qingyun When the latest config is used, the `label_embedding` problem disappeared. Thanks for helping. I checked with the paper and the source code published by the authors, and it seems...

I met a new problem, it happens in the middle of training, such as 2000 iterations. I'm trying to fix it. I notice that it happens when `dn_cls_scores` is None,...

@Li-Qingyun Only set `filter_empty_gt=True` is not enough, I also set `allow_negative_crop=False`. Image without any bbox ground-truth will cause this problem.

> @Li-Qingyun Only set `filter_empty_gt=True` is not enough, I also set `allow_negative_crop=False`. Image without any bbox ground-truth will cause this problem. My model is still in training and I can...

@Li-Qingyun My last box AP is 48.4 and 0.6 points lower than the official repo (49.0 box AP), I suspect skipping negative samples may cause some performance degradation.

> > 我遇到了一个新问题，它发生在训练的中间，比如 2000 次迭代。我正在尝试修复它。 我注意到它在`dn_cls_scores`为 None 时发生，因此 loss_dict 中的键在 GPU 之间是不同的。 > > ``` > > AssertionError: loss log variables are different across GPUs! > > rank 2 len(log_vars):...

Hello, thanks for your attention. I read the Colab notebook you wrote. The bug is that it can't find the pre-trained model. You need to download the pre-trained model first...

Hi, can you add `CUDA_LAUNCH_BLOCKING=1` to the `dist_train.sh` and provide me with the new log? Like this: ``` #!/usr/bin/env bash CONFIG=$1 GPUS=$2 PORT=${PORT:-29300} #PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \ CUDA_LAUNCH_BLOCKING=1 python -m torch.distributed.launch...