search

DonDominic / MP2Net

The official code of our TGRS'24 paper MP2Net: Mask Propagation and Motion Prediction Network for Multi-Object Tracking in Satellite Videos
MIT License
17 stars 0 forks source link

Details about calculating FPS #7

Closed lebron-2016 closed 5 months ago

lebron-2016 commented 5 months ago

Dear author,

How did you calculate the FPS? What is the relevant experimental configuration?

Looking forward to your reply.

Thanks.

DonDominic commented 5 months ago

@lebron-2016 We simply tested it on SatVideoDT and calculated the average frame rate under the overall sequences as FPS.

Hope this finds you well.

lebron-2016 commented 5 months ago

@lebron-2016 We simply tested it on SatVideoDT and calculated the average frame rate under the overall sequences as FPS.

Hope this finds you well.

Hi, do you mean the method of dividing the total test time by the number of test images? I used a single TITAN RTX to perform inference on the SatVideoDT dataset using mp2net_resnet. The timing is shown in the following figure:

image

The calculated FPS is 1.2, which is quite different from the 2.5 in the paper. Could you provide more details?

In addition, how is FLOPS calculated in the paper?

Thanks for your help!

DonDominic commented 5 months ago

@lebron-2016 Please check if there are other programs occupying computing resources and whether the hard disk used is SSD. The calculation process of FLOPs is included in stNet.py.

Hope this finds you well.

lebron-2016 commented 5 months ago

@lebron-2016 Please check if there are other programs occupying computing resources and whether the hard disk used is SSD. The calculation process of FLOPs is included in stNet.py.

Hope this finds you well.

Got it! Thanks!!