search

Sense-X / Co-DETR

[ICCV 2023] DETRs with Collaborative Hybrid Assignments Training
MIT License
913 stars 94 forks source link

summary of model #137

Open prsbsvrn opened 3 months ago

prsbsvrn commented 3 months ago

Hello, I need detailed information about the model, and I would like to use torchinfo's summary function. Could you advise on where I should incorporate the following code? from torchinfo import summary input_size = (3, 224, 224) # Example input size (C, H, W) for a typical image print("Summary of model") summary(model, input_size)

I have already attempted to integrate it into train.py within both the tools and apis folders of the mmdet directory. Your guidance would be greatly appreciated.

TempleX98 commented 2 months ago

Some operations (e.g., deformable attention) in the model do not support model summary.

prsbsvrn commented 2 months ago

Thanks for your response. How can I learn about flops and the impact of each part of the model on flops? Can you guide me?

TempleX98 commented 2 months ago

The total FLOPs of Co-DINO-5scale is the same as DINO-5scale since our training technique does not introduce additional inference costs. The FLOPs of the DINO model is presented in Table 1 of its paper here.

prsbsvrn commented 2 months ago

Thank you, what about Co Deformable DETR?