qidian213/ranked_person_reid

Person reID

ranked_person_reid's Issues

STN

Why add STN before the base model? Is there the performance improvement? And why?

训练不了

会出现这个问题，是和C++有关的吗？

How about Person ReID using Example Weighting (OSM&CAA, RLL, ICE, IMAE, DM)?

Dear ReID experts:

If possible, I sincerely recommend you trying our example weighting methods.
The reasons:
(1) ReID datasets may have noisy observations or labels, and sample imbalance. I find that example weighting is a good approach for addressing these challenges.
(2) We can discuss and work together to make it if there is a chance.
(3) I am not an expert in ReID, which makes it harder for me to make it alone.

1. ReID using RLL: https://github.com/Qidian213/Ranked_Person_ReID
2. ReID using OSM and CAA: Deep Metric Learning by Online Soft Mining and Class-Aware Attention.

• Code of us: https://github.com/XinshaoAmosWang/OSM_CAA_WeightedContrastiveLoss
• Code by third party: https://github.com/ppriyank/-Online-Soft-Mining-and-Class-Aware-Attention-Pytorch

3. ReID on MARS using IMAE: https://github.com/XinshaoAmosWang/Improving-Mean-Absolute-Error-against-CCE
4. ReID on MARS using DM: https://github.com/XinshaoAmosWang/DerivativeManipulation

Related Papers:
OSM and CAA: https://arxiv.org/abs/1811.01459 (Robust metric learning & classification)
RLL: https://arxiv.org/abs/1903.03238 (metric learning)
IMAE: https://arxiv.org/pdf/1903.12141.pdf (robust classification)
DM: https://arxiv.org/pdf/1905.11233.pdf (robust classification and general example weighting)

Hi @Qidian213
I put this post here for open discussion and collaboration.
If it is not okay for me, please let me know. Otherwise, I appreciate it greatly.
Thanks.

How to reproduce your best result?

Hi, I was trying to reproduce your best result (mAP=88.7%) by simply runing

python tools/train.py --config_file='configs/softmax_ranked.yml' DATASETS.NAMES "('market1501')" 

but I got a much worse result at mAP=86.5%, with 2% margin to claimed result. Moreover, it's even lower than using triplet loss. What I can do to get your claimed result?

model	mAP	loss
r50-ibn-a(my result)	87.1%	softmax + triplet
r50-ibn-a(my result)	86.5%	softmax + ranked
r50-ibn-a(your result)	88.7%	softmax + ranked

Thank you!

Couldn't unpack resnet50_ibn_a.pth.tar!

Can you give me an uncompressed model?

Person ReID using Ranked List Loss, IMAE, OSM and CAA, ....

So glad to hear from you, looking forward to discussion and sharing with you:

1. ReID using RLL: https://github.com/Qidian213/Ranked_Person_ReID

2. ReID results using RLL will be shown: https://github.com/XinshaoAmosWang/Ranked-List-Loss-for-DML

3. ReID using OSM and CAA: https://github.com/ppriyank/-Online-Soft-Mining-and-Class-Aware-Attention-Pytorch
   https://arxiv.org/abs/1811.01459

4. ReID on MARS using IMAE: https://github.com/XinshaoAmosWang/Improving-Mean-Absolute-Error-against-CCE

Related Papers:
OSM and CAA: https://arxiv.org/abs/1811.01459 (metric learning & classification)
RLL: https://arxiv.org/abs/1903.03238 (metric learning)
IMAE: https://arxiv.org/pdf/1903.12141.pdf (classification)

准确率问题

您好，请问为啥用crank_loss训练的结果反而比rank_loss的结果差呢？您也一样的嘛？并且我的准确率始终只有map83%。r1=93%左右在market1501数据集上。

你好，请问一下为什么会有两种结果？最后一个是怎么得到的呢？

resnet_ibn_a.py error

i set resnet_ibn_a path with _C.MODEL.NAME = 'resnet50_ibn_a'
but show
param_dict = model_weight['state_dict']
KeyError: 'state_dict'

thank your works with hope reply.

FileNotFoundError: [Errno 2] No such file or directory: 'modeling/resnet50_ibn_a.pth.tar'

您好！打扰了!
resnet50_ibn_a.pth.tar 我在网上找不到这个文件，您能提供一下吗？

tensorrt

Can't use tensorrt ，because of the IBN layer?thanks

The result cannot be reproduced.

Hi, I can't reproduce your work on the market1501 dataset. Specifically, I use resnet50 as the backbone and without rerank, I can only get results of mAP 84.1% / rank-1 93.5%, which is far from the results ( mAP 87.2% / rank-1 95%) given in your report.

My run script is as follows：
python3 tools/train.py DATASETS.NAMES "('market1501')" MODEL.NAME "('resnet50')"

have bug

Frames Per Second

Hello! Can you please tell, how much is the frames per second value