jiangweibeta / mlic Goto Github PK

作者您好，我想请教一下您文章里那么多的方法的结果都是从哪里得到的，我只能从公开的资源上获取，那些未公开的您都是邮件联系得到的吗？不知道您是否愿意将其开源。谢谢。期待您的回复

Update 2024-04-08
I upload the training log when the lambda is 0.0250. The model is trained on 4 GPU cards with ddp support.

I fix the LatentResidualPrediction and SynthesisTransform, you should use LatentResidualPredictionFix and SynthesisTransformFix. The parameter number becomes 83.5M. The modification leads to no performance drop.

It would be great if you share pretrained models after this update.

fine tuning时，报如下错误，是对训练集的图片尺寸有要求吗？
24-04-10 10:36:36.398 - INFO: Learning rate: 0.0001
Traceback (most recent call last):
File "/home/zhanghh/work/JiangWeibeta/MLIC/MLIC++/playground/train.py", line 162, in
main()
File "/home/zhanghh/work/JiangWeibeta/MLIC/MLIC++/playground/train.py", line 124, in main
current_step = train_one_epoch(
File "/home/zhanghh/work/JiangWeibeta/MLIC/MLIC++/utils/training.py", line 24, in train_one_epoch
optimizer.step()
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/lr_scheduler.py", line 75, in wrapper
return wrapped(*args, **kwargs)
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/optimizer.py", line 385, in wrapper
out = func(*args, **kwargs)
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/optimizer.py", line 76, in _use_grad
ret = func(self, *args, **kwargs)
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/adam.py", line 166, in step
adam(
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/adam.py", line 316, in adam
func(params,
File "/home/zhanghh/work/JiangWeibeta/MLIC/venv/lib/python3.10/site-packages/torch/optim/adam.py", line 520, in _multi_tensor_adam
torch.foreach_lerp(device_exp_avgs, device_grads, 1 - beta1)
RuntimeError: The size of tensor a (32) must match the size of tensor b (192) at non-singleton dimension 0

有看到 dataload部分
train_transforms = transforms.Compose(
[transforms.RandomCrop(args.patch_size), transforms.ToTensor()]
)
对图像进行裁剪，理论上只要数据集中图片尺寸大于 args.patch_size 就行，对吧？

非常感谢你提供的源代码和测试数据，我想问你一下其他论文中的参数量和FLops是怎么计算的，你有什么建议么

请问可以上传MLIC和MLIC+的代码吗？

ImportError: Cannot Import 'ste_round' from 'compressai.ops'.
I cannot find 'ste_round' in the GitHub Pages(https://interdigitalinc.github.io/CompressAI/).

Hello, author. Thank you for your work. I would like to ask, in the training set you selected, to ensure fairness, do all the compared methods need to be retrained on the training set you selected?

img.resize((new_width, new_height), Image.ANTIALIAS)
请问new_width和new_height是设置成多少呢？