DEIM

DEIM copied to clipboard

I trained the model using my own dataset, with the detection categories being person, car, and bike, and the training set size is 1 million samples. However, compared to YOLOv5,...

MrPanda007

AssertionError assert (boxes1[:, 2:] >= boxes1[:, :2]).all()

12

Hello，I have a bug **randomly** when i train by custome dataset, can you give me some idea to fix this bug😭😭? there is log: ```bash Epoch: [48] [ 0/112] eta:...

Sunburst7

“nan" problem in training custom dataset

16

Epoch: [18] [3500/4384] eta: 0:08:08 lr: 0.000003 loss: 31.5320 (30.7038) loss_vfl: 0.6807 (0.6671) loss_bbox: 0.1400 (0.1503) loss_giou: 0.4664 (0.4601) loss_fgl: 1.0752 (1.0601) loss_vfl_aux_0: 0.7646 (0.7457) loss_bbox_aux_0: 0.1571 (0.1619) loss_giou_aux_0: 0.5047...

ohmyhair151263

训练三轮以后就出现这个问题

3

rank : 0 (local_rank: 0) exitcode : 1 (pid: 16677) error_file: traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html

1zaa

torch.topk(scores.flatten(1), self.num_top_queries, dim=-1)，这行表示什么意思

2

我是负责做模型推理的，我的理解是应该每个query只选择分数最高的那个类别， 如果进行flatten，选择top 300，那么一个query可能选择多个类别，即一个框对应多个类别，这好像没有意义。

demuxin

有什么办法得到混淆矩阵？

skydoubleone

请问为什么我在训练Deim-dfine-n-coco模型的时候，GPU利用率只有15%，而CPU利用率达到了500%

2301979087

resume断点续训，接着训练loss从31增加到46，epoch训练时长从3小时增加5小时

1

Houseqin

mal loss 上升

2

训练初期mal loss上升的原因是什么？ 

Shu-Ang

The error when reached 90th training epoch

2

I use 'CUDA_VISIBLE_DEVICES=0 torchrun --master_port=7777 --nproc_per_node=1 train.py -c configs/deim_dfine/deim_hgnetv2_n_GC10.yml --use-amp --seed=0' to train the model. When I reached my 90th round of training epoch, the error occurs as follow: ...

Livier18