Zhuofan Zong comments

Results 111 comments of


                                            Zhuofan Zong

Error while trying to train model for coco. Error states "torchvision.models.segmentation.image_demo" not found

torch 1.11.0+cu113 torchaudio 0.11.0+cu113 torchvision 0.12.0+cu113 These are my torch packages. Do you use cuda 11.3?

Error while trying to train model for coco. Error states "torchvision.models.segmentation.image_demo" not found

Can you provide more error logs?

PyTorch Fine Tuning

This is not supported yet. Community PR is welcome.

No module named 'einops'

Thanks for your feedback. I will add this dependency to requirements.

推理出来的特别多的检测框

config里的test_cfg是一个list，分别对应detr head和多个辅助头的test setting。你这里显示的score_thr都是加到了辅助头的config，可以试试加到test_cfg[0]里

Training Stuck at "mmcv - INFO - load checkpoint from torchvision path: torchvision://resnet50"

This means you are downloading the resnet50 checkpoint from torchvision. Please check your internet connection.

ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 1428) of binary: /opt/conda/bin/python

Please check the file `mmdet/models/losses/cross_entropy_loss.py` in your environment's mmdetection package. Is this file the same as [the implementation](https://github.com/Sense-X/Co-DETR/blob/28a60116d9bcdd22a876a358b160a27efee785ec/mmdet/models/losses/cross_entropy_loss.py) of this repo?

在coco数据集上只使用12个epochs就足够了吗

训练epoch数量取决于你的dataset大小，coco的数据量级是10w左右，可以根据这个预估然后通过实验找到最佳epoch设置

Which model in the HuggingFace repo includes Objects365 pretraining?

The O365 model is here: https://huggingface.co/zongzhuofan/co-detr-vit-large-objects365