Models: Pretrained resnet 18/34 model?

[meijieru]

System information

• What is the top-level directory of the model you are using: slim
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 14.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 1.2
• Bazel version (if compiling from source): N/A
• CUDA/cuDNN version: CUDA 8.0, cuDNN 5.1
• GPU model and memory: GTX TITAN X PASCAL, 13GB
• Exact command to reproduce: N/A

Describe the problem

Feature Request

Could you please provide pretrained resnet18/34 on imagenet?

Thanks very much!

[jingjingwang]

Also, could you provide pre-trained models using VGG as the feature extractor? I understand that they're not the best ones but still, many people would be interested in using them.

[reedwm]

It appears we don't support training a resnet 18/34 model either, in the Slim resnet_v1.py file.

@sguada any plans to create pretrained models for resnet18/34?

[Cospel]

ResNet was moved to official directory so it will be great if someone with good gpu can provide some pretrained weights for all the configurations: 18,34,50,101,152,200

https://github.com/tensorflow/models/tree/master/official/resnet

[wt-huang]

Closing this as different sizes can be configured through --resnet_size in resnet.py.

[JoeyCarson]

How is this issue solved by just configuring a different size? The ask was to provide a pretrained ResNet-18 model. As of today, ResNet-50 is the only one available. Perhaps we can specify a smaller size and those variables will be able to be warm started from the ResNet-50 model, but that isn’t the same as providing as pretrained ResNet-18. Am I wrong?

[abrarmatin]

Can we expect a pre-trained Resnet 34 model to be available? Has there been an update on this?

[davidlrobinson]

Tensorflow really should include ResNet-18 and -34 by default.