Ouail Bendidi

the training code is not yet published by the authors of deep_sort, you can find more information [here](https://github.com/nwojke/deep_sort/issues/15) and [here](https://github.com/nwojke/deep_sort/issues/7) you can use sort alternatively

i'm working on adding the new layers of yolov3 to darkflow, I already added the `shortcut` and `upsampling` layers, only missing the `yolo` layer , so expect it soon if...

you can change the GPU Flag in YOLO : `FLAGS.gpuName = '/gpu:0'` by default it's on GPU 0 , but you can set it to another GPU (for example /gpu:1)...

try using `with tf.device("GPU:1"):` before calling the encoder I'm sorry I don't have the required hardware so can't really help much

`output[0]` is a tensor with shape `[xxxx , NUM_CLASSES]` , as it is the output of a softmax layer (confidence of each class) , you could simply add an argmax...

division by 10 is because we run the FCN in a loop over 10 image so to get time per image you divide by 10

yes the goal is to get the "time by frame" or average inference time as you said it , and if you inverse it you will get "frame per time"=fps

I'm using pool5 as the fcn input, and got it working at 3 fps (I didn't do any optimization or input queues) the results were note that much different from...

i'm using a GTX 1080 for inference and running from python script too, i did an average on 100 predictions and i got around 2.7 fps (using tf 1.0.1 with...

I've redone the test with only measuring inference (sess.run) (in the last tests I calculated the whole process from reading the image and prepossessing to doing the predictions) and I've...