xyq7 / invcompress Goto Github PK
View Code? Open in Web Editor NEW[ACMMM 2021 Oral] Enhanced Invertible Encoding for Learned Image Compression
License: Apache License 2.0
[ACMMM 2021 Oral] Enhanced Invertible Encoding for Learned Image Compression
License: Apache License 2.0
Hello,
Great work and awesome code. Thank you very much for sharing.
I see that total 20,745 images are present in Flickr dataset out of which 200 random images are taken away for validation, and then generated patches of size 256*256.
If I use the generate patch script, it results huge number of patches. How many patches are present in your training dataset?
Regards
Priyanka
Hi Yueqi, thanks for your great work!
I am wondering about a small bug of the network architecture.
According to the paper, for the SSIM models, the para N
for the first two lambda values is 128, while that for the last three lambda values is 192.
However, according to the code, the para N
for the first four lambda values is 128, while that for the last four lambda values is 192.
I want to make sure that if I want to train the all five SSIM models, my training code should be:
python examples/train.py ../ -exp exp_ssim_q1_01 -m invcompress -d ../data/flickr2w --epochs 600 -lr 1e-4 --batch-size 8 -n 8 --cuda --gpu_id 7 -q 1 --lambda 6 --metrics ms-ssim --save --seed 7
python examples/train.py ../ -exp exp_ssim_q2_01 -m invcompress -d ../data/flickr2w --epochs 600 -lr 1e-4 --batch-size 8 -n 8 --cuda --gpu_id 6 -q 1 --lambda 12 --metrics ms-ssim --save --seed 7
python examples/train.py ../ -exp exp_ssim_q3_01 -m invcompress -d ../data/flickr2w --epochs 600 -lr 1e-4 --batch-size 8 -n 8 --cuda --gpu_id 5 -q 5 --lambda 40 --metrics ms-ssim --save --seed 7
python examples/train.py ../ -exp exp_ssim_q4_01 -m invcompress -d ../data/flickr2w --epochs 600 -lr 1e-4 --batch-size 8 -n 8 --cuda --gpu_id 4 -q 5 --lambda 120 --metrics ms-ssim --save --seed 7
python examples/train.py ../ -exp exp_ssim_q5_01 -m invcompress -d ../data/flickr2w --epochs 600 -lr 1e-4 --batch-size 8 -n 8 --cuda --gpu_id 3 -q 5 --lambda 220 --metrics ms-ssim --save --seed 7
Is it correct? Thank you.
How can you use the examples/codec.py file to evaluate your model?
Hi, I try to load the pre-trained model and fine-tune it in my own dataset. However, there exists a bug in loading the model:
InvCompress/codes/examples/train.py
Line 444 in 17deb0f
In this line, there has an KeyError: 'state_dict'.
It seems that the model is not saved as train.py in line 501:
save_checkpoint(
{
"epoch": epoch + 1,
"state_dict": net.state_dict(),
"loss": loss,
"optimizer": optimizer.state_dict(),
"aux_optimizer": aux_optimizer.state_dict(),
"lr_scheduler": lr_scheduler.state_dict(),
},
is_best,
os.path.join('../experiments', args.experiment, 'checkpoints', "checkpoint_%03d.pth.tar" % (epoch + 1))
)
And the size of different pictures at the same compression level is the same
Hi! Thanks for your great work!
During measuring PSNR values of test images, is there any difference between them?
Since the quantization method of 'forward()' is slightly different from that of 'compress()',
I'm wondering how much the difference occurs.
Hi authors,
I found that the provided link only contains MSE models. How can I download the MS-SSIM models?
Thanks
Excellent work! When would you like to release the code?
Thank you for your codes. However when I try to use the pre-trained model (exp_02_mse_q2), some errors occurred. Other pre-trained model including (q1, q3,q4,q5) can work normally.
Thanks for your last reply! When I try to train the model under higher compression (nearing 0.06), I get a poor performance. So do you have the pretrained model under higher compression ?
Hi,
Great work ! I'm currently using it in my phd research project.
If I understand the code correctly, the context model (ctx_p) is significantly slowing down the entropy coding/decoding computation...
Is there any way to make this much faster ?
Thanks.
Can you provide the results of opt-msssim as well?
Thanks
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.