RuntimeError: The size of tensor a (12828) must match the size of tensor b (3000) at non-singleton dimension 1
SJavad opened this issue · 3 comments
SJavad commented
Hi 🖐
I Trained an SSD model with --resolution=640 and when I want to convert the .pth model to .onnx, it gives me an error:
locations[..., :2] * center_variance * priors[..., 2:] + priors[..., :2],
RuntimeError: The size of tensor a (12828) must match the size of tensor b (3000) at non-singleton dimension 1when I want to convert the model to 300*300 size, it is OK and converts successfully...
but when I set the --width=640 and --height=640, this error happened.
exported command:
Namespace(net='ssd-mobilenet', input='', output='', labels='labels.txt', width=640, height=640, batch_size=1, model_dir='.\\models\\GED\\')
=> running on device cuda:0
=> found best checkpoint with loss 2.568583369255066 (.\models\GED\mb1-ssd-Epoch-6-Loss-2.568583369255066.pth)
creating network: ssd-mobilenet
num classes: 3
=> loading checkpoint: .\models\GED\mb1-ssd-Epoch-6-Loss-2.568583369255066.pth
=> exporting model to ONNX...SJavad commented
A LOT of open issues and... 🤯
no answer...
Hmmm 🤔
Interesting... 🧐
Zalways commented
hey！i met the same problem with you! have you solved this problem?looking forward to your reply!