What about the result of your experiments?

InnovArul / personreid_sequential_rl

An attempt to replicate the paper "Multi-shot Pedestrian Re-identification via Sequential Decision Making (CVPR2018)"

10 stars 4 forks source link

What about the result of your experiments? #1

Closed LaviLiu closed 5 years ago

LaviLiu commented 5 years ago

Is the reinforcement policy is power according your experiments？ Could you share your result for us?

InnovArul commented 5 years ago

I was not able to reproduce the base paper's results. The base architecture results (Alexnet + Softmax +/ Triplet) in the paper did not coincide with my finding (the base paper had a very low number). Hence, the results of DQN from my experiments are way-off from the base paper.

The report I wrote for this project is here : https://github.com/InnovArul/personreid_sequential_rl/blob/master/docs/project_report.pdf

InnovArul commented 5 years ago

I hope you found solutions to your questions. I am closing this Github issue. Pl. reopen if you still have questions.