Is it possible to train Resnet-50/Resnet-110 in TitanX Pascal 12GB?

[john1...@gmail.com]

Hello all, I am using ResNet-50 and Titan X pascal 12GB. I would like to know is it possible to train the network in the GPU card? I means the GPU memory. Thank all

[Hieu Do Trung]

I can train ResNet-50 on my GTX-960 with only 4GB of VRAM, using batch size of 8, so you absolutely can train it on GTX Titan X with 12GB.
As for ResNet-110 I'm not sure, but given the above number (batch size of 8 on 4GB GPU), I guess you can train ResNet-110 on a 12GB GPU.

[Johnson Mark]

Thanks Hieu. 

Is possible to share your training Resnet-50?

Best Regards,

John

Vào 10:59:00 UTC+9 Thứ Sáu, ngày 23 tháng 12 năm 2016, Hieu Do Trung đã viết:

[Hieu Do Trung]

Sure! These are the files that some guys created and I just modified it a bit.

In the train_val-resnet50-template.prototxt file, you need to change #train_batch and #test_batch to your desired values (as my GPU only has 4GB, I use 8 and 1, respectively).
Also change the "/train_mean.binaryproto", "/train_lmdb", "/val_mean.binaryproto", "/val_lmdb",  #num_output, and some paths in the solver file to your own values.

[Rubina Rashid]

Hi
Any one help for getting the pretrained Resnet-50 model for finetuning....

[Bartosz Ludwiczuk]

Try sometimes asking google:  https://github.com/KaimingHe/deep-residual-networks#models

[Z.L]

Thank you for your sharing those files. Could you please also share us train batch and step_size, max_iter? By the way, What is your final test accuracy top 1 and top 5? 

Zhe