Li-Chongyi
Questions for heatmap visualization
Hello, Author.
Thanks for your fantastic work on zero-reference low-light image enhancement.
I've recently been trying to reproduce the visualization in Python, as shown in Figure 3, to see a pixel-by-pixel curve parameter map for attempting the visualization is to verify that the Zero-DCE model is trained as intended. I have a few questions along the way.
According to your paper and codes, I believe that I can get the pixel-wise curve parameter map through the DCE-Net, organized as (Batch_size x 24 x Height x Width) for eight iterations (Three R, G, B color channels x 8).
For visualization, First, Referring to the DCE-Net output illustration in your paper, I divided the data in the form of (24 x Height x Width) into (8 x Height x Width) slices, mapping them to the colors of
(0:8, H, W) = R (8:16, H, W) = G (16:24, H, W) = B
respectively.
After that, I averaged the color map data through the iterations, (8 x H x W) -> (1 x H x W) of each color, and normalized the values to the range of [0, 1].
I think this process's result should be the same as the heatmap shown in Figure 3 in the paper. Still, the result heatmap images are some kind of values that are highly related to the brightness of the enhanced image. (although different values are obtained for the R, G, B regions).
Would you tell me how I can get the heatmap-like visualized parameter map in your paper using Python?
