Model Inference Taking ".jpg" files List

[AliWaheed]

Hi, I would like to run "run_tracking.py" on Numpy files rather than passing the list of ".jpgs" I followed the inference session and saw that you are providing jpg list right to your model for inference. Can you please guide a bit on how can I run inference on Numpy images.

[bilylee]

For the best efficiency, it is recommended to use jpeg images. The reason is that TensorFlow supports directly reading jpeg images from the file system.

To your question, one choice is to convert NumPy files to jpeg images, then run the tracking script. Nevertheless, it is also possible to use Numpy files with some modifications.

1. modify the input pipeline of the computation graph to use Numpy array instead of jpeg filenames

https://github.com/bilylee/SiamFC-TensorFlow/blob/f572dca95f2b3b2861f54de467259753428e468c/inference/inference_wrapper.py#L80-L84

2. change the feed_dict to use Numpy array instead of filename

https://github.com/bilylee/SiamFC-TensorFlow/blob/f572dca95f2b3b2861f54de467259753428e468c/inference/inference_wrapper.py#L226

https://github.com/bilylee/SiamFC-TensorFlow/blob/f572dca95f2b3b2861f54de467259753428e468c/inference/inference_wrapper.py#L237

3. change the input_feed during tracking

https://github.com/bilylee/SiamFC-TensorFlow/blob/f572dca95f2b3b2861f54de467259753428e468c/inference/tracker.py#L57

https://github.com/bilylee/SiamFC-TensorFlow/blob/f572dca95f2b3b2861f54de467259753428e468c/inference/tracker.py#L78

[AliWaheed]

Thank you for such quick reply. I will check it out and report back