DAB-DETR

DAB-DETR copied to clipboard

How to use the swin-transformer as the backbone for DAB-DETR

5

Thank you for help!

yuan738

Swim-Transformer Pre-trained Weights

4

Dear Shilong, Thanks for the amazing work again! In your codes, it supports swim-transformer as the backbone, but I could not find its ImageNet pre-trained weights. Could you please kindly...

ZhangGongjie

nvcc fatal : Unsupported gpu architecture 'compute_86'

1

When I setup deformable multi head attention, report the error as below: > nvcc fatal : Unsupported gpu architecture 'compute_86' > Traceback (most recent call last): > File "/data/queenie_liu/anaconda3/envs/mmTrans/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line...

fengsky401

Compiling CUDA operators error

4

when try to compile CUDA operator I get this errors: C:/Users/Ali/DAB-DETR-main/models/dab_deformable_detr/ops/src\cuda/ms_deform_im2col_cuda.cuh(97): error: identifier "__floorf" is undefined in device code C:/Users/Ali/DAB-DETR-main/models/dab_deformable_detr/ops/src\cuda/ms_deform_im2col_cuda.cuh(98): error: calling a __host__ function("__floorf") from a __device__ function("ms_deform_attn_col2im_bilinear ")...

Aliweka2020

Hi, I added `--no_aux_loss` to the eval script command.

qritive-ola

Temperature tuning and positional embeddings for DAB-DETR

3

In the paper, there is a section saying the optimal temperature for positional embedding is 20 in your model. However, this [line](https://github.com/IDEA-opensource/DAB-DETR/blob/309f6ad92af7a62d7732c1bdf1e0c7a69a7bdaef/models/DAB_DETR/transformer.py#L43) under gen_sineembed_for_position indicates that a value of 10000...

YellowPig-zp

The meaning of num_feature_levels.

2

Thanks for your great and detailed code! In `dab_deformable_detr.py`, is this part about dealing with the insufficient channel of features (for Deformable DETR)? > https://github.com/IDEA-opensource/DAB-DETR/blob/309f6ad92af7a62d7732c1bdf1e0c7a69a7bdaef/models/dab_deformable_detr/dab_deformable_detr.py#L169-L181 It seems that the last...

TsingWei

PostProcess

1

Is it possible that multiple labels of a query are topks during post-processing? How to solve this situation?

Twxwx

Why modulating attention by w&h works?

1

I have some doubts on line https://github.com/IDEA-opensource/DAB-DETR/blob/main/models/DAB_DETR/transformer.py#L242 . ``` refHW_cond = self.ref_anchor_head(output).sigmoid() # nq, bs, 2 ``` This line asks the model to learn absolute value of w, h from...

SupetZYK

@SlongLiu Nick work ! The entries of [tgt_bbox ](https://github.com/IDEA-Research/DAB-DETR/blob/main/models/DAB_DETR/matcher.py#L83)are already [x0, y0, x1, y1] after boxes[:, 2:] += boxes[:, :2] [in this line](https://github.com/IDEA-Research/DAB-DETR/blob/main/datasets/coco.py#L93). So I want to know the underlying...

qiulesun