Manu Mathew

After cloning this repository, in your bash prompt, change directory to the caffe-jacinto-models/scripts folder, run the following command: ./train_image_object_detection.sh The prototxt files will be generated in the folder caffe-jacinto-models/scripts/training For...

Can you try the base code https://github.com/NVIDIA/caffe with the same flags and let me know if it works?

Hi, please note the url that I mentioned. https://github.com/NVIDIA/caffe Clone this repository, then git checkout caffe-0.17 Then try to build using the flags that you mentioned.

Okay. Can you then file an issue in that repository? https://github.com/NVIDIA/caffe Let us know if you find a solution for this issue.

Please see: https://github.com/tidsp/caffe-jacinto/issues/30

@dainHandle Do you get any error while running the scripts? My expectation is that the old links continue to work for now. Note: I will be on vacation for a...

Can you try the following: 1. Temporarily switch to teh the git branch caffe-0.16 for both caffe-jacinto and caffe-jacinto models. 2. Then clean build caffe-jacinto 3. see if your training...

You can visualize your net train.prototxt and test.prototxt with netscope tool. The error is saying that 5 outputs are expected, but only 3 are obtained.

This issue is now fixed. You can pull the latest code. Basically, do not set chop_num_heads to 1. Instead comment out those lines that set it to 1 in train_image_object_detection.sh:...

What we are doing is simulating quantization and measuring how much error would occur if you this kind of fixed point result. Thus you know what degradation to expect when...