search

bardiadoosti / HOPE

Source code of CVPR 2020 paper, "HOPE-Net: A Graph-based Model for Hand-Object Pose Estimation"
270 stars 58 forks source link

Will the pretrained model be provided? #1

Closed billyzju closed 4 years ago

billyzju commented 4 years ago

Could you please provide the pretrained model? Thanks!

bardiadoosti commented 4 years ago

Sure the pretrained model for egocentric view will be provided in the next few days.

bardiadoosti commented 4 years ago

Pretrained model is added in the README file instruction.

billyzju commented 4 years ago

Pretrained model is added in the README file instruction.

OK, that's very nice of you. Is that possible to test on my own image by the way? Thank you.

bardiadoosti commented 4 years ago

Yes, make the .npy file for your dataset and test it on that particular dataset.

billyzju commented 4 years ago

Yes, make the .npy file for your dataset and test it on that particular dataset.

To test on my own dataset, do I need to provide the camera external/internal references? Like here: https://github.com/bardiadoosti/HOPE/blob/8ea544d36847dfd660e5c409695376595c89920f/datasets/fhad/make_data.py#L51

bardiadoosti commented 4 years ago

For 3D annotation you need these matrices. But if you just want to test the model on a 2D annotated dataset, put a place holder for 3D labels (maybe a zero vector) and manually remove the 3D term for the loss. That will just calculate the 2D error for you.

ghost commented 4 years ago

Hi @bardiadoosti. Thank you for sharing your code. Here is my inference code:

` model.load_state_dict(torch.load(args.pretrained_model))

image_path = os.path.join(cwd, 'data/images') image_name = 'frame_029.jpg' image_file = os.path.join(image_path, image_name) image = Image.open(image_file) img = transform(image) img = img.unsqueeze(0)

inputs = img

inputs = Variable(inputs)

if use_cuda and torch.cuda.is_available(): inputs = inputs.float().cuda(device=args.gpu_number[0]) with torch.no_grad(): outputs2d_init, outputs2d, outputs3d = model(inputs)

outputs2d = outputs2d.cpu().numpy() outputs2d_init = outputs2d_init.cpu().numpy() print(outputs2d.shape) f,ax = plt.subplots(1,1, figsize=(10, 10)) ax.imshow(image) ax.scatter(outputs2d[:,:,0], outputs2d[:,:,1]) save_image_file = os.path.join(image_path, '{}_out.jpg'.format(image_name.split('.')[0])) plt.savefig(save_image_file)`

And the result is shown in the image below. Do you have any suggestion?

frame_029_out