search

mousecpn / Joint-Bilateral-Learning

An unofficial implementation of Joint Bilateral Learning for Real-time Universal photorealistic Style Transfer
MIT License
66 stars 23 forks source link

Can you offer the pre-trained model weights? #7

Open liupgd opened 3 years ago

liupgd commented 3 years ago

Hi, I'm interested with this model. But I can't run the test without your pretrained weights. Appreciated if you can offer the pretrained weights. Thx. My email: liupgd@foxmail.com

mousecpn commented 3 years ago

Hi, I'm interested with this model. But I can't run the test without your pretrained weights. Appreciated if you can offer the pretrained weights. Thx. My email: liupgd@foxmail.com

Sorry, I didn't train successfully on COCO dataset.

liupgd commented 3 years ago

Hi, I'm interested with this model. But I can't run the test without your pretrained weights. Appreciated if you can offer the pretrained weights. Thx. My email: liupgd@foxmail.com

Sorry, I didn't train successfully on COCO dataset.

I trained on COCO2017, but the result is far away from the paper. I used 100k images, same data operation with the paper. I checked your code and the BGU paper and the JBL paper. I think there is almost no fault with your code.
Here is my result trained 29 epochs:

individualImage

All the test images are searched with google. The first two results are not as good as the paper. I don't know why. Maybe, the training dataset also matters.

oferidan1 commented 3 years ago

hi liupgd, mousecpn , can you please upload your pre-trained model? I understand its not perfect, but it would be great if you can share it. thanks in advanced.

liupgd commented 3 years ago

hi liupgd, mousecpn , can you please upload your pre-trained model? I understand its not perfect, but it would be great if you can share it. thanks in advanced.

Hi @oferidan1 , I modified the training to pytorch-lightning. I've merged my code to here: https://github.com/liupgd/Joint-Bilateral-Learning checkpoint: https://drive.google.com/file/d/1BH5RyvRcwm0VC75iEcp2NLe4V7imyAn2/view?usp=sharing some parameters: image

It has been so long, I can't remember all details now. All above is what I can find out from my computer. Hope it will help you.

oferidan1 commented 3 years ago

thanks alot, liupgd.

liupgd commented 3 years ago

好像是的，好久前的事了，记得不太清了。-------- 原始邮件 --------发件人： martin_chen @.>日期： 2021年6月17日周四 19:43收件人： mousecpn/Joint-Bilateral-Learning @.>抄送： liupgd @.>, Author @.>主 题： Re: [mousecpn/Joint-Bilateral-Learning] Can you offer the pre-trained model weights? (#7)

hi liupgd, mousecpn , can you please upload your pre-trained model?

I understand its not perfect, but it would be great if you can share it.

thanks in advanced.

Hi @oferidan1 , I modified the training to pytorch-lightning. I've merged my code to here: https://github.com/liupgd/Joint-Bilateral-Learning

checkpoint: https://drive.google.com/file/d/1BH5RyvRcwm0VC75iEcp2NLe4V7imyAn2/view?usp=sharing

some parameters:

It has been so long, I can't remember all details now. All above is what I can find out from my computer. Hope it will help you.

你好 能讲中文吧

我想问一下数据集是怎么做的 是随机在coco2017中选取风格图片和内容图片各50000张吗

—You are receiving this because you authored the thread.Reply to this email directly, view it on GitHub, or unsubscribe.

zhanghongyong123456 commented 2 years ago

hi liupgd, mousecpn , can you please upload your pre-trained model? I understand its not perfect, but it would be great if you can share it. thanks in advanced.

Hi @oferidan1 , I modified the training to pytorch-lightning. I've merged my code to here: https://github.com/liupgd/Joint-Bilateral-Learning checkpoint: https://drive.google.com/file/d/1BH5RyvRcwm0VC75iEcp2NLe4V7imyAn2/view?usp=sharing some parameters: image

It has been so long, I can't remember all details now. All above is what I can find out from my computer. Hope it will help you.

I use yours pre model ,but get error,i dont know why ,can you help me ? my test cmd

python test.py --cont_img_path /home/zhy/Projects/RealHM/vendor_testing_2/pexels-aleksandar-pasaric-2115367.jpg --style_img_path /home/zhy/Projects/RealHM/vendor_testing_2/pexels-aleksandar-pasaric-2115367_fore.jpg --model_checkpoint ./checkpoints/JBL-epoch=73.ckpt

and error:

image

liupgd commented 2 years ago

Hi,@zhanghongyong123456 if you are using both my pytorch-lightning version of code and the pretrained model. please just use main_pl.py for training and testing purpose. The pytorch-lightning models are different from the original ones and can't be loaded by pytorch directly. The main_pl.py' s arguments has shown how to do the test. Hope this help you.

zhanghongyong123456 commented 2 years ago

Hi,@zhanghongyong123456 if you are using both my pytorch-lightning version of code and the pretrained model. please just use main_pl.py for training and testing purpose. The pytorch-lightning models are different from the original ones and can't be loaded by pytorch directly. The main_pl.py' s arguments has shown how to do the test. Hope this help you.

yes,I want to know how the parameters are configured, I only have one GPU, I don't know where to pass the test image (content and style)，only add checkpoint path

image

zhanghongyong123456 commented 2 years ago

Hi,@zhanghongyong123456 if you are using both my pytorch-lightning version of code and the pretrained model. please just use main_pl.py for training and testing purpose. The pytorch-lightning models are different from the original ones and can't be loaded by pytorch directly. The main_pl.py' s arguments has shown how to do the test. Hope this help you.

Hello, I have successfully tested the model according to your prompt, I have two questions looking forward to your reply, 1. Can I achieve 4K real-time 1080 on the computer（in paper,I know mobile phone 4k real time ?） 2. How to set the high resolution (2K 4K), the result I get is 512 Resolution, if it can be any resolution, set it there, I am not too familiar with the pytorch_lightning library