Manu Mathew

One question (for me to uderstand). When did you last pull the code from caffe-jacinto?

It probably means that your original training did use pretrained model correctly. You can probably compare the logs of your working training and not working to check this.

The inference script is here: https://github.com/tidsp/caffe-jacinto-models/blob/caffe-0.17/scripts/infer_cityscapes_segmentation.sh Example prototxts with quantization are in: trained/image_segmentation/cityscapes5_jsegnet21v2/test_quantize All you have to do is add quantize: true to your prototxt file. Not that this is...

You may also want to merge the batch norm weights to convolution weights using caffe optimize command before running the inference with quantization.

Please see these links: https://github.com/NVIDIA/caffe/issues/549 https://github.com/NVIDIA/caffe#useful-notes

1. Center loss The definition of your Center Loss layer is in the style of https://github.com/BVLC/caffe However, this repository is based on https://github.com/NVIDIA/caffe Please look at the definition of layers...

Hi, I think I did not understand the problem that you are facing. Specifically: Yow wrote: "But I can use 2 gpus to train." If you can use 2 gpus...

You have to install blas. I am not sure if its is called cblas or openblas. Try installing both. You may also have to install atlas. In Ubuntu, I find...

Some clarifications: 1. Tried out some of my custom configurations - and it trains well and produces good results. However I have not tried to reproduce the original mAP of...

jdetnet21v2 is small because its targeted for low power embedded devices. What I meant is that if you train with the original VGG based configuration published at https://github.com/weiliu89/caffe/tree/ssd, you should...