How to get the same result in Market1501
Closed this issue · 5 comments
Hi,
Using the code in the repo and Market1501 dataset you should be able to reproduce the results.
If you could please give more information on the problems you encountered I may be able to help you solve them.
Dear @mikwieczorek,
I am trying to reproduce the result from your pre-trained weight in Market1501 dataset. Following here are the results:
top-k, Rank-1 :93.9%
top-k, Rank-5 :98.2%
top-k, Rank-10 :98.9%
top-k, Rank-20 :99.4%
top-k, Rank-50 :99.6%
mAP: 0.8538947533255574
It seems like lower than in the paper. I followed the instruction in the repo. Everything is smooth, except for some minor errors related to version. After seeing the results, I also encountered the error as below:
AttributeError: 'Trainer' object has no attribute 'accelerator_backend'
Exception ignored in: <function tqdm.del at 0x7f759a7614c0>
Traceback (most recent call last):
File "/usr/local/lib/python3.8/dist-packages/tqdm/std.py", line 1152, in del
File "/usr/local/lib/python3.8/dist-packages/tqdm/std.py", line 1306, in close
File "/usr/local/lib/python3.8/dist-packages/tqdm/std.py", line 1499, in display
File "/usr/local/lib/python3.8/dist-packages/tqdm/std.py", line 1155, in str
File "/usr/local/lib/python3.8/dist-packages/tqdm/std.py", line 1457, in format_dict
TypeError: cannot unpack non-iterable NoneType object
Do you have any advice for me regarding to this case? I am still trying to get the same result as you indicated. Hopefully I can have your opinion. Thank you in advance!
Hi @petertran1811,
The results you posted here are the results without using centroids for evaluation. If you run the test as stated in the Readme file, there are two evaluations happening. 1) stage uses no-centroids evaluation and it gives the results you provided. 2) stage using centroids for evaluation and this should give the results from the paper.
If you encountered the Error the 2nd stage did not start probably.
You can run centroids evaluation directly adding MODEL.USE_CENTROIDS True to the command you used.
@petertran1811 Hello, have you tried?