search

fredzzhang / pvic

[ICCV'23] Official PyTorch implementation for paper "Exploring Predicate Visual Context in Detecting Human-Object Interactions"
BSD 3-Clause "New" or "Revised" License
67 stars 8 forks source link

About pre-trained weight of object detector #33

Closed xz-123-new closed 1 year ago

xz-123-new commented 1 year ago

Thanks for your great work! I have one question for your pretrained weight of object detector when training ViC-H-Defm-DETR-SwinL on HICO-DET, the weight of h-detr: h-defm-detr-swinL-dp0-mqs-lft-iter-2stg-hicodet.pth, how should i choose it from the model zoo in https://github.com/HDETR/H-Deformable-DETR, because there are many models satisfies backbone swin-large, or can i just randomly pick one of them? Thanks !

xz-123-new commented 1 year ago

Empirically i find the shape mismatch because of the object class number differs from COCO and HICO-DET, and i wonder where could i load the pretrained weight for HICO-DET, thanks!

fredzzhang commented 1 year ago

Hi @xz-123-new,

The weights are slightly different because we fine-tuned H-DETR on HICO-DET, and it only has 80 object classes as opposed to 91 in COCO.

The weights will be release once I'm back from ICCV.

Fred.

KPeng9510 commented 1 year ago

Hi,

I am also interested in your pretrained weights on HICO--DET for H-DETR. May I ask when you are planning to upload them :)?

fredzzhang commented 1 year ago

Hi @KPeng9510 and @xz-123-new,

I just realised I already uploaded the weights here a few months ago. Apologies for the wait.

Fred.

fredzzhang commented 1 year ago

Detailed documentation has now been added here.

Fred.