mathsshen / dasiamrpnwithofflinetraining Goto Github PK

Python 100.00%

dasiamrpnwithofflinetraining's Introduction

Shen Li

Institute of Data Science, National University of Singapore

I'm a third-year PhD student of National University of Singapore under the supervision of Prof. Bryan Hooi. My research interests include deep generative models, face recognition and visual object detection and tracking. Prior to this, I received my M.S. from University of Chinese Academy of Sciences where I was advised by Prof. Xilin Chen. I received my B.S. in computer science from Beijing Jiaotong University.

News

September, 2021. I received Google PhD fellowship 2021 on the machine learning track.

August, 2019. I will be joining IDS-NUS as a PhD student under the supervision of Dr. Bryan Hooi.

June, 2018. I received my M.S. from University of Chinese Academy of Sciences.

July, 2018. I received IEEE ICME 2018 Platinum Best Paper Award.

Publications

Ailin Deng, Shen Li, Miao Xiong, Zhirui Chen, Bryan Hooi. Trust but Verify: Using Self-Supervised Probing to Improve Trustworthiness. European Conference on Computer Vision 2022 (ECCV).

Yuqiong Qi, Yang Hu, Haibin Wu, Shen Li, Xiaochun Ye, Dongrui Fan. A synergistic reinforcement learning-based framework design in driving automation. Computers and Electrical Engineering Volume 101, July 2022, 107989.

Shen Li, Bryan Hooi. Neural PCA for Flow-Based Representation Learning. International Joint Conference on Artificial Intelligence (IJCAI) 2022.

Miao Xiong, Shen Li, Wenjie Feng, Jihai Zhang, Ailin Deng, Bryan Hooi. Birds of a Feather Trust Together: Knowing When to Trust a Classifier via Adaptive Neighborhood Aggregation. Transactions on Machine Learning Research (TMLR) 2022.

Yuqiong Qi, Yang Hu, Haibin Wu, Shen Li, Haiyu Mao, Xiaochun Ye, Dongrui Fan, Ninghui Sun. Tackling Variabilities in Autonomous Driving. arXiv preprint, 2021.

Shen Li, Jianqing Xu, Xiaqing Xu, Pengcheng Shen, Shaoxin Li, Bryan Hooi. Spherical Confidence Learning for Face Recognition. 2021 IEEE International Conference on CVPR (Oral, acceptance rate top 4.3%) .

Shen Li, Bryan Hooi, Gim Hee Lee. Identifying through Flows for Recovering Latent Representations. 2020 IEEE International Conference on Learning Representations (ICLR2020).

Shen Li, Xiaqing Xu, Bingpeng Ma, Hong Chang, Xilin Chen. Learning Hidden States for Visual Tracking. Preprint, 2018.

Shen Li, Bingpeng Ma, Hong Chang, Shiguang Shan, Xilin Chen. Continuity-Discrimination Convolutional Neural Network for Visual Object Tracking. 2018 IEEE International Conference on Multimedia and Expo (ICME2018, Oral, Platinum Best Paper Award).

Youfang Lin, Shen Li, Sujie Liu, Yuchang Chen. An Efficient Approach to Mobile Robot Motion Planning in Dynamically Unknown Environments. IEEE ICARCV 2014: 1764-1770 (Oral).

dasiamrpnwithofflinetraining's People

Contributors

Stargazers

Watchers

Forkers

perfectzh saintlogos1234 viody zj360202 tangtangchx linzhi123 alexadlu jianjiangg skyneta shiyongde changhezhai kongbia caijiahao daiwc spinoza honglinchu swan2015 isoil666 minyuedayuan

dasiamrpnwithofflinetraining's Issues

Test not working

Thank you very much for your work!
But when I run the model with incremental learning using SiamRPNBig.model provided by the author of the paper, it seems not work. I edit the net.py file to use the trained model without changing the structure of the network. Would you please give me some suggestions on my problem.

About training sample

could you please share a demo of sample? I'm a little confused of its format. Thank you.

Can you share me your way of contact?

Can you share me your way of contact? I really need your hlep about this project. Thanks!

illegal memory access was encountered

File "train.py", line 260, in
mask = torch.from_numpy(mask).to(DEVICE)
RuntimeError: CUDA error: an illegal memory access was encountered

How to convert into onnx thank you

view(-1) and nan loss

Traceback (most recent call last):
File "train.py", line 245, in
reg_loss = reg_loss_fn(reg_output[i], torch.from_numpy(rpn_bbox_targets).to(DEVICE), torch.from_numpy(rpn_bbox_inside_weights).to(DEVICE), torch.from_numpy(rpn_bbox_outside_weights).to(DEVICE))
File "train.py", line 62, in smooth_l1_loss
loss_box = loss_box.view(-1).sum(0) / N
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

how to solve this problem? I am not familiar with pytorch, I am not sure whether my revision to this code was correct. I changed as follows:

loss_box = loss_box.view(-1).sum(0) / N
+loss_box = torch.reshape(loss_box, (-1,)).sum(0) / N

but the loss become nan like
Epoch0 Iter1 --- total_loss: nan, cls_loss: nan, reg_loss: 0.1331
17%|██████████████████████████████ | 2/12 [00:03<00:22, 2.22s/it]Epoch0 Iter2 --- total_loss: nan, cls_loss: nan, reg_loss: 0.6188
25%|█████████████████████████████████████████████ | 3/12 [00:05<00:17, 1.98s/it]Epoch0 Iter3 --- total_loss: nan, cls_loss: nan, reg_loss: 1.5328
33%|████████████████████████████████████████████████████████████ | 4/12 [00:06<00:14, 1.81s/it]Epoch0 Iter4 --- total_loss: nan, cls_loss: nan, reg_loss: 2.0786

I got overwhelmed!!

Is this repo sharing the same parameters with foolwood version？

I use the same pretrained model SiamRPNBIG.model in https://github.com/foolwood/DaSiamRPN to run this repo. (after motified some code). However, it's extremely slow(can't even achieve 1 fps). I run foolwood version with these parameters very normally.

Was the model trained with OTB instead of ILSVRC VID?

about the loss

Hello, thank u for ur great work!!!! Thanks a lot.
After modifying some code, the train.py can run successfully.
But the loss is very strange, as follow:
0%| | 0/8 [00:00<?, ?it/s]Epoch0 Iter0 --- total_loss: nan, cls_loss: nan, reg_loss: 0.6299
12%|█▎ | 1/8 [00:02<00:14, 2.01s/it]Epoch0 Iter1 --- total_loss: nan, cls_loss: nan, reg_loss: 1.7844
25%|██▌ | 2/8 [00:03<00:09, 1.61s/it]Epoch0 Iter2 --- total_loss: nan, cls_loss: nan, reg_loss: 34.9781
38%|███▊ | 3/8 [00:04<00:07, 1.48s/it]Epoch0 Iter3 --- total_loss: nan, cls_loss: nan, reg_loss: 238.0343
50%|█████ | 4/8 [00:05<00:05, 1.41s/it]Epoch0 Iter4 --- total_loss: nan, cls_loss: nan, reg_loss: 236.5256
62%|██████▎ | 5/8 [00:06<00:04, 1.37s/it]Epoch0 Iter5 --- total_loss: nan, cls_loss: nan, reg_loss: 70.5485
75%|███████▌ | 6/8 [00:08<00:02, 1.35s/it]Epoch0 Iter6 --- total_loss: nan, cls_loss: nan, reg_loss: 113.4333
88%|████████▊ | 7/8 [00:09<00:01, 1.33s/it]Epoch0 Iter7 --- total_loss: nan, cls_loss: nan, reg_loss: 56.8303
100%|██████████| 8/8 [00:10<00:00, 1.25s/it]
Saving model...

Have u met it before? Thanks.

About distractor-aware training implementation

Hi, MathsXDC:
Thank you for your great work! I've just read your implementation, and I'm having trouble finding the implementation of distractor-aware training part in your code. I briefly summarized my questions as below:

How do you sample negative pairs in your code?
Did you test your performance on the vot benchmark?

It would be great if you could enlighten me on above questions. Thank you very much!

some training questions

when running the code, the cls_loss will be nan. I think it is because of the 'mask' in train.py can be zero. But I am not sure. And another question is that when running the code, there will be a error "cuda runtime error: an illegal memory access was encountered". hoping to hear from you soon. thank you very much