This repo is a Tensorflow implementation of paper Full Resolution Image Compression with Recurrent Neural Networks. This repo also contains a trained Residual LSTM model (saved in save/
) on a small dataset (doesn't include dataset). Here is one example shown in the paper
Original image:
Reconstruction:
Training, put training data in imgs/
folder and run following code for default setting. The model parameter can be modified in model.py
. Running time comparison of original and reconstructed images can be seen in eval/
. Model file is saved in save/model
.
python train.py
Encoding
python encode.py --model save/model --input kodim05.png --iters 10 --output compressed.npz
Decoding
python decode.py --model save/model --input compressed.npz --output compressed.png
Evaluation, code from Tensorflow's official repo
python msssim.py -o kodim05.png -c compressed.png
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。