search

nwojke / deep_sort

Simple Online Realtime Tracking with a Deep Association Metric
GNU General Public License v3.0
5.16k stars 1.45k forks source link

time between detection and updata of tracker #127

Open louxy126 opened 5 years ago

louxy126 commented 5 years ago

Hi, I am doing some researches on the time consuming between detection and updata of tracker，I combined deep_sort with faster-rcnn and I calculate the time consuming between the detection outcome is given (boxs = detect_one_image(frame, pred_func)) and tracker is updated tracker.update(detections). I found that the first frame took 10 times as long as the latter here _run_in_batches( lambda x: self.session.run(self.output_var, feed_dict=x), {self.input_var: data_x}, out, batch_size) I am a little confused about it. Is this because session is called for the first time?

nwojke commented 5 years ago

This is likely due to TensorFlow. To fix the timing, you could make a single call on random data before you actually process any "real" data.

FernandoCalatayud commented 5 years ago

Hi @louxy126 . I'm trying to use yolov3 and deep SORT, as in yolov3 detects every 30th frame, and the rest is taken care by deep SORT. Did you managed to do that with faster-rcnn?