Using model on own data set. about vid2vid HOT 8 CLOSED

You can take a look at the provided example dataset. You dataset should be in the same format.

from vid2vid.

I tried looking into the example data sets. Would I have the same architecture if I want a image to image transformation? Would i just need Train_A, Train_B and then Test_A and Test_B, and then I just need to specify data-root?

from vid2vid.

If you just want image-to-image translation instead of videos, you can just use this repo: https://github.com/NVIDIA/pix2pixHD/

from vid2vid.

We are currently using pix2pix as well, but want to see if vid2vid will yield better results since our pictures are a sequence of images. I saw that in the sample data sets, there are more than just .png file. I haven't got a chance to run vid2vid yet, but will we be able to run vid2vid with just train_A and train_B folders with just pictures?

from vid2vid.

Yes, make sure the images in the two folders are in corresponding order (i.e. first image in train_A corresponds to first image in train_B), and it should work.

from vid2vid.

Also try to see if I can run vid2vid on my own dataset. Are the sequence sub-folders (e.g., seq0001 for cityscapes dataset) necessary? Or I can simply group all image sequences under the train_A folder?

from vid2vid.

I tried looking into the example data sets. Would I have the same architecture if I want a image to image transformation? Would i just need Train_A, Train_B and then Test_A and Test_B, and then I just need to specify data-root?

I am trying to replicate the results of the pose model. I trained the model on colab. I have a sequence of images and openpose keypoints JSON file corresponding to each frame, that I need to provide to get the generated video for my use case... But I don't see Train_A,Train_B folders. Since I am trying it in colab, the .py PyTorch variables wont be accessible from colab command line. Please help me here. Thanks in advance.

from vid2vid.

@tcwang0509 @yuanzhou15
At present, I am also doing image to image translation. I have tried pix2pix and pix2pixHD, but since my image is a continuous video frame, I also want to try the vid2vid method. I want to ask the following questions:

1. what is the test effect of this method?
2. Thecontinue_ train command seems to be unavailable. It needs to be retrained every time. Have you ever encountered this?
3. if use_real_img command is used, the first frame is In folder test_B? Must the number of images in folder test_B correspond to that in folder test_A?

Looking forward to your reply!!!

from vid2vid.