search

makbari7 / DSSLIC

DSSLIC: Deep Semantic Segmentation-based Layered Image Compression
44 stars 19 forks source link

result tested on Kodak dataset #3

Open ChenWei007HB opened 4 years ago

ChenWei007HB commented 4 years ago

Hi, author, here is the issue I met when I test the model provided by you from the link in readme files on the Kodak dataset, the result shows can not surpass BPG method when using the residual coded by BPG, and lossless codec method FLIF, it still has a big gap between DSSLIC model and BPG, I've tried many ways to improve the result, but failed, I hope you can help me to figure it out, thanks.

iwa-shi commented 2 years ago

Hi, @ChenWei007HB . I'm not the author, but I think there are some problems with this paper and code. First, the bitrate calculation is probably wrong. According to the line 135 in evaluation code/main.m, I think total_bits is divided by height * width * channel, not height * width. That means the result may be 1/3 of the correct bitrate. Second, the authors set the color_space of BPG to RGB. In most papers, YCbCr is used rather than RGB. I tried to plot the actual RD curve on Kodak dataset. In this figure, the proposed method performs worse than BPG. I would like to get your feedback. Thanks. ScreenShot 2021-11-03 16 48 20

Echo-cwb commented 2 years ago

@iwa-shi Hello, iwa-shi. I also downloaded the code and tried to run it, but encountered a problem with the dataset import: AssertionError: /DSSLIC/ADE20K/train_label is not a valid directory. I would like to know how the ADE20k data format of this code is, thank you very much for your help

adityak55 commented 1 year ago

Hello Author,I've tried to download the dataset and it showed the error "The requested URL was not found on this server".Could you please upload the DataSet