Pretrained models about posecontrast HOT 3 CLOSED

Yes, the input to viewpoint estimation network is cropped image, where the bounding boxes can be given by ground truth or detected by an object detection model.

For class-agnostic object detection, you can use the class-agnostic MaskRCNN trained on MS-COCO in this repo for detecting cars in StanfordCars images.

from posecontrast.

Hi @ofir-dl ,

Thanks for your interests! I have updated the README, you can now find the command to download the model trained on Pascal3D here

from posecontrast.

Thanks it's working :)

I would like to run inference on StanfordCars dataset for pose estimation.
Seems that your dataset already has bounding boxes corresponding to the objects within its images.

Does it mean I should first detect the cars and then use your model for the pose estimation?
Do you plan to add inference support to this repo?

Thank you so much!

from posecontrast.