Training on real images

[zhangy76]

Hi,

I have setup the environments and run the training code "python train_smpl.py --cfg configs/orig_cliff.yaml". But errors occur in finding the images in Human3.6M and MPI-INF-3DHP.

May I ask how you save the Human3.6M and MPI-INF-3DHP video data to image data? Is that starting from 1 with writing step 5 (seems not...)?

Thanks, Yufei

[pixelite1201]

Hello, Could you tell me what error occurs? Also it is not very clear what you mean by starting from 1 with writing step 5.

[zhangy76]

Hi,

Thank you for the quick reply! The errors are failing to load the images from Human3.6M and MPI-INF-3DHP.

After I downloaded the video data from the website, I will need to extract and image frames from it. For a video, I save the images every 5 frames following the sequence order. The image name of the first frame is 000001.jpg. But it seems the images are not extracted like what I did. So I am wondering how you extract the images?

Yufei

[pixelite1201]

Sorry for the delay in reply. We have extracted all images into a folder and used every 5th frame. Here is a screenshot of one of the h3.6m folder after extracting images.

[zhangy76]

Hi,

Thanks for the reply. Would you mind sharing the preprocessing code for these datasets? I am considering to avoid any possible confusions.

Thanks, Yufei

[pixelite1201]

Sorry, we already have the data extracted in image format on our server. So, I didn't use any code for preprocessing. If you still face issues with the preprocessing, please share your error and I will try to help you.

[chojinie]

@pixelite1201 Hi, author. I am trying to train the CLIFF on real image sets. And have a similar issue as zhangy76's problem.

i fix H36m issue with 50FPS. But, i wanna know about mpi_inf_3dhp dataset, What is the frame rate (FPS) at which images should be extracted from the 3DHP dataset?