# Video Frame Synthesis using Deep Voxel Flow
[[Project]](https://liuziwei7.github.io/projects/VoxelFlow) [[Paper]](https://arxiv.org/abs/1702.02463)
-<img src='./misc/demo.gif' width=640>
+<img src='./misc/demo.gif' width=720>
## Overview
`Deep Voxel Flow (DVF)` is the author's re-implementation of the video frame synthesizer described in:
"Video Frame Synthesis using Deep Voxel Flow"
-[Ziwei Liu](https://liuziwei7.github.io/), [Raymond A. Yeh](http://www.isle.illinois.edu/~yeh17/), [Xiaoou Tang](http://www.ie.cuhk.edu.hk/people/xotang.shtml), [Yiming Liu](http://bitstream9.me/), [Aseem Agarwala](http://www.agarwala.org/)
+[Ziwei Liu](https://liuziwei7.github.io/), [Raymond A. Yeh](http://www.isle.illinois.edu/~yeh17/), [Xiaoou Tang](http://www.ie.cuhk.edu.hk/people/xotang.shtml), [Yiming Liu](http://bitstream9.me/), [Aseem Agarwala](http://www.agarwala.org/) (CUHK & UIUC & Google)
In International Conference on Computer Vision (ICCV) 2017, Oral Presentation
<img src='./misc/demo_teaser.jpg' width=800>
``` bash
python voxel_flow_train.py --subset=test
```
+* Run the evaluation script:
+``` bash
+matlab eval_voxelflow.m
+```
## License and Citation
The use of this software is RESTRICTED to **non-commercial research and educational purposes**.
--- /dev/null
+clear;clc;
+
+num_img = length(subdir);
+flag_valid = zeros(1, num_img, 'single');
+mat_psnr = zeros(1, num_img, 'single');
+mat_ssim = zeros(1, num_img, 'single');
+
+for id_img = 1:num_img
+
+ dir_img_cur = [dir_data, subdir(id_img).name, '/'];
+
+ % read images
+ img_pred = imread([dir_img_cur, 'pred_01.png']);
+ img_target = imread([dir_img_cur, 'target_01.png']);
+ img_prev = imread([dir_img_cur, 'frame_00.png']);
+ img_next = imread([dir_img_cur, 'frame_01.png']);
+
+ img_pred_ycbcr = rgb2ycbcr(uint8(img_pred));
+ img_target_ycbcr = rgb2ycbcr(uint8(img_target));
+
+ img_pred_gray = img_pred_ycbcr(:, :, 1);
+ img_target_gray = img_target_ycbcr(:, :, 1);
+
+ img_pred = single(img_pred);
+ img_target = single(img_target);
+ img_prev = single(img_prev);
+
+ img_pred_gray = single(img_pred_gray);x
+ img_target_gray = single(img_target_gray);
+
+ % check validity
+ if sum(mask_flow(:)) > 0
+
+ flag_valid(id_img) = 1;
+
+ img_pred_mask = repmat(mask_flow, [1, 1, 3]) .* img_pred;
+ img_target_mask = repmat(mask_flow, [1, 1, 3]) .* img_target;
+
+ mse = sum((img_pred_mask(:) - img_target_mask(:)).^2) ./ (3 .* sum(mask_flow(:)));
+ psnr_cur = 20.0 .* log10(255.0) - 10.0 .* log10(mse);
+
+ [ssim_cur, ~] = ssim(img_pred_gray, img_target_gray);
+
+ mat_psnr(id_img) = psnr_cur;
+ mat_ssim(id_img) = ssim_cur;
+
+ end
+
+ disp(['Processing Img ', num2str(id_img), '...']);
+
+end
+
+flag_valid(find(mat_psnr == inf)) = 0;
+mat_psnr(find(mat_psnr == inf)) = 0;
+
+mean_psnr = sum(flag_valid .* mat_psnr) ./ sum(flag_valid)
+mean_ssim = sum(flag_valid .* mat_ssim) ./ sum(flag_valid)