When I used this web(jetson-inference/pytorch-collect.md at master · dusty-nv/jetson-inference · GitHub) to train googlenet network, I have added photos to all categories, but the feature recognition rate has not changed. I don’t why.
Hi @andy8902, do you mean the accuracy during training with PyTorch has not increased? How many images did you add, and how many images total are in your train dataset? And how many classes do you have.
If the accuracy improved during training, but not inferencing with imagenet/TensorRT program, make sure you delete the .engine file in your models folder after you ran onnx_export.py. It’s possible that TensorRT is running an old engine (i.e. from the previous model you trained)
When I incresed pictures on old model training, the picture recognition rate has not changed.
Then I changed new model , the picture recognition rate has changed.
Hi,
Please noticed that you will need to re-train the model first.
The re-training process will update the weights of model based on the new database.
Thanks.