search

dBeker / Faster-RCNN-TensorFlow-Python3

Tensorflow Faster R-CNN for Windows/Linux and Python 3 (3.5/3.6/3.7)
MIT License
613 stars 329 forks source link

How to replace VGG with resnet #79

Closed S0soo closed 5 years ago

S0soo commented 5 years ago

I tried to modify the training file, but there were still many mistakes. Did anyone succeed

morpheusthewhite commented 5 years ago

I need more information to help you.

What's the error printed in the console?

S0soo commented 5 years ago

I need more information to help you.

What's the error printed in the console?

File "E:\faster-rcnn\lib\nets\resnet_v1.py", line 27, in weight_decay=cfg.TRAIN.WEIGHT_DECAY, AttributeError: module 'lib.config.config' has no attribute 'TRAIN' II don't know how to deal with this error, if you know, please help me, thanks a lot

morpheusthewhite commented 5 years ago

Try to replace the following line (27)

                     weight_decay=cfg.TRAIN.WEIGHT_DECAY,

with 

                     weight_decay=cfg.FLAGS.weight_decay,
morpheusthewhite commented 5 years ago

After some tests I found out that resnet needs some effort to work, now it's not supported.

You can either create a pull request or wait for @dBeker's answer

S0soo commented 5 years ago

Thank you very much, I understand, and I find cfg.resnet and cfg.pooling_size to be unmanageable

S0soo commented 5 years ago

https://github.com/endernewton/tf-faster-rcnn I have successfully trained according to this, and I will have a test after the training. Thank you very much for your advice

morpheusthewhite commented 5 years ago

You're welcome

chenjjjjuu commented 5 years ago

https://github.com/endernewton/tf-faster-rcnn I have successfully trained according to this, and I will have a test after the training. Thank you very much for your advice

hi !Can you used resnet to trained your data on fasterrcnn by this link?I have been to try it but faild!

S0soo commented 5 years ago

yes，you can download files that need to be replaced in my GitHub

chenjjjjuu commented 5 years ago

yes，you can download files that need to be replaced in my GitHub

ok,I'll try it. thanks

S0soo commented 5 years ago

yes，you can download files that need to be replaced in my GitHub

ok,I'll try it. thanks

You're welcome. I wish you success.

chenjjjjuu commented 5 years ago

yes，you can download files that need to be replaced in my GitHub

ok,I'll try it. thanks

You're welcome. I wish

yes，you can download files that need to be replaced in my GitHub

ok,I'll try it. thanks

You're welcome. I wish you success.

excuse me. I try your script in linux not windows.there have an error in datasets/pascal_voc.py in line 34,what's the mean of "self.path = r'E:\faster-rcnn\labels_10_JYsf.txt'",and where is my labels_10_JYsf.txt?

S0soo commented 5 years ago

This is your class name, a .txt file

chenjjjjuu commented 5 years ago

This is your class name, a .txt file

can you show me your .txt for example please? I don't konw which form should the .txt be. I just have one class named flower.

S0soo commented 5 years ago

like this image

chenjjjjuu commented 5 years ago

and where can download resnet_v1_50.cpkg?

S0soo commented 5 years ago

image

chenjjjjuu commented 5 years ago

image

i find it. thanks

chenjjjjuu commented 5 years ago

image

image

like this

like this image

datasets like this? VOCdevkit2007 -VOC2007 -Annotations Imagesets JEPGImages

S0soo commented 5 years ago

yes，

chenjjjjuu commented 5 years ago

https://github.com/endernewton/tf-faster-rcnn I have successfully trained according to this, and I will have a test after the training. Thank you very much for your advice

could you send me your file plesae, I can't run this in resnet finally.

S0soo commented 5 years ago

https://github.com/endernewton/tf-faster-rcnn I have successfully trained according to this, and I will have a test after the training. Thank you very much for your advice

could you send me your file plesae, I can't run this in resnet finally.

I'm sorry, I can't send you the code I used, because I added a lot of content on this basis, the modified code is very different from the previous one. But I can try to help you solve some problems.

chenjjjjuu commented 5 years ago

thank you so much and what's your mean 'cfg.resnet and cfg.pooling_size to be unmanageable'.

(https://github.com/S0soo/tf-faster-rcnn-resnet)This code can run in linux?

S0soo commented 5 years ago

This runs in win. To run in linux, use https://github.com/endernewton/tf-faster-rcnn

chenjjjjuu commented 5 years ago

ok I'll try ------------------ 原始邮件 ------------------ 发件人: "YLF"notifications@github.com 发送时间: 2019年10月4日(星期五) 下午5:37 收件人: "dBeker/Faster-RCNN-TensorFlow-Python3"Faster-RCNN-TensorFlow-Python3@noreply.github.com; 抄送: "chenjjjjuu"1540358700@qq.com;"Comment"comment@noreply.github.com; 主题: Re: [dBeker/Faster-RCNN-TensorFlow-Python3] How to replace VGG withresnet (#79)

This runs in win. To run in linux, use https://github.com/endernewton/tf-faster-rcnn

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

S0soo commented 5 years ago

'cfg.resnet and cfg.pooling_size to be unmanageable' means if you want to use resnet, you need to add some extra hyperparameters

S0soo commented 5 years ago

For this demo, if you have any questions, you can add my QQ, 1572128279.