This is a source code written in CUDA/C++ to train ResNet-34 on the ImageNet dataset.
We refered these excellent base codes.
[1] https://github.com/pjreddie/darknet/tree/a3714d0a2bf92c3dea84c4bea65b2b0c64dbc6b1
[2] https://github.com/PacktPublishing/Learn-CUDA-Programming
The training result is as follows:
The learning rate and batch size are not fully tuned.
However, the trained ResNet-34 model reached at the 71.51% accuracy.
It is comparable to the result from Darknet's ResNet-34 (72.4%).