Re-training ResNet-18 Model

924971015 · December 9, 2020, 7:34am

I am learning the tutorials of jetson-inference. Now I am stuck in this step of Re-training ResNet-18 Model.When I run the train.py, I encounter the problem below the picture.I am really looking forward your reply, thank you very much.

AastaLLL · December 10, 2020, 7:16am

Hi,

It seems that you are meeting a similar problem of the below issue:

github.com/dusty-nv/jetson-inference

Runtime error - train.py

opened 07:43PM - 29 Nov 20 UTC

DJT777

Hey dusty, I am having a runtime error that I am unable to solve on my own. W…ould you take a look at this output in the terminal when I am attempting to do the training part of the tutorial? ``` dylan@tx1:/media/dylan/SD/jetson-inference/python/training/classification$ sudo python3 train.py --model-dir=models/cat_dog data/cat_dog Use GPU: 0 for training => dataset classes: 2 ['cat', 'dog'] => using pre-trained model 'resnet18' => reshaped ResNet fully-connected layer with: Linear(in_features=512, out_features=2, bias=True) Traceback (most recent call last): File "train.py", line 506, in <module> main() File "train.py", line 135, in main main_worker(args.gpu, ngpus_per_node, args) File "train.py", line 277, in main_worker train(train_loader, model, criterion, optimizer, epoch, num_classes, args) File "train.py", line 334, in train acc1, acc5 = accuracy(output, target, topk=(1, min(5, num_classes))) File "train.py", line 500, in accuracy correct_k = correct[:k].view(-1).float().sum(0, keepdim=True) RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. ```

Please give it a check.
Thanks.

dusty_nv · December 10, 2020, 7:37pm

Hi @924971015, are you using PyTorch 1.7? I found this issue began occurring because of some updates in PyTorch 1.7 - I just fixed it in the classification training script in commit 15c7d5

If you are using PyTorch 1.7, re-clone the jetson-inference repo and the classification training should work now.

The updated code should still work on previous versions of PyTorch.

924971015 · December 13, 2020, 2:55am

Thank you very much , I have solved this problem.

924971015 · December 13, 2020, 2:59am

Is your problem identical with mine, could you describe your problem in details?

924971015 · December 13, 2020, 3:03am

the problem have been solved by dusty , please re-clone the jetson-inference repo and the classification training should work now.