changgyhub / reconet-torch

Re-implementation by liulai@

reconet-torch

This repository contains a PyTorch implementation of the ReCoNet paper. It use a lot of code from safwankdb

Contributors:

• Mohd Safwan
• Kushagra Juneja
• Saksham Khandelwal
• safwankdb

dataset

we use the MPI Sintel Dataset. The dataset contains flow fields, motion boundaries, unmatched regions, and image sequences. The image sequences are rendered with different levels of difficulty.

Train

python train.py --phase train --path mpidataset --style-name=style_images/vanGogh.jpg

video

python video_cv2.py --video-name 'avi/output_style.avi' --mode video_style --save-directory trained_models --model-name model.pth

nodes

You can use optim.Adam(), it results bubble artifacts in the following red boxes. I used optim.Adamax() in my code.