Hi All,
I want to implement real-time image classification using Jetson TX2. I have my own data-set comprising of 3 different classes. I initially trained them using Tensorflow on Inception V3 architecture and it I have a good accuracy but I realized that I can’t straight way deploy the model on Jetson TX2 and there are 2 ways of making tensorflow models work on jetson.
1)Train in DIGITS from scratch or retrain the pre-trained models(like Alexnet, GoogleNet) and download the model snapshot to deploy the model in Jetson.
The issue i faced is that I can’t upload my pre-trained model because it says some json file is missing and I thought i could retrain it on inception V3 architecture using DIGITS but the DIGITS model repo doesn’t have inception V3 model.
2)The second way is to freeze the graph and convert the model into TensorRt compatible one and run on Jetson. but I wasn’t able to do that either and I am getting some decode error(UnoCodeDecodeError: ‘utf-8’) I couldn’t proceed any further on it.
The files that was generated after training in tensorflow using inception V3 architecture is;
1.output_graph.pb
2.output_labels.txt
3._retrain_checkpoints.meta
4._retrain_checkpoint.data-00000-of-00001
5._retrain_checkpoint.index
6.checkpoint
Please can anyone advise me on this and tell me know to proceed further?
With regards
Sri