iamanorange / dsslic Goto Github PK

Dear lamanorange:
From the Checkpoints folder, I found there exists a pre-trained ADE20k model, but the link is invalid, I wonder if you still save this model. Thank you very much!

Dear Iamanorange:
Hello, I would like to ask if you have a model file that has been trained and can be used to test the effect directly. I want to see the test effect of the entire model, but I have not successfully run train.py. Thank you for your help

I tried to run this code locally but encountered a problem with the dataset import: AssertionError: /DSSLIC/ADE20K/train_label is not a valid directory. The ADE20k data is downloaded from the website you suggested. I would like to know what the ADE20k data format of this code is, thank you very much for your help.

Dear All,

Unfortunately I am not possible to reproduce results shown in the linked paper, both for pre-trained network and for the self-trained network.

With the pre-trained networks I get the same results as in the paper on the ADE20K and Kodak datasets. On the other hand, on cityscapes, the results I get are the same for the DSSLIC network. However, the results for classical codecs are much better than what is presented in the paper.

When I train the networks myself, the results are far from those in the paper. Even if I continue with training of the pre-trained network with a "super small" learning rate for a few epochs, it seems that the performance of the DSSLIC network is drastically worsened by this small perturbations of the network parameters.

As this strange behaviour of the DSSLIC codec kept me puzzled for a few months, any tips that could help me reproduce results from the paper would be much appreciated. This would also be beneficial for the reputation of the paper, as some authors started reporting worse DSSLIC's results in their papers.

Dear Iamanorange:
I have run your code in the Colab, but it is difficult to import training data, I would like to know if there is a new version of the code or code in .ipynb format on your side. Thank you for your generous help