zhanght021 / rpt Goto Github PK

您好，我对您的代码挺感兴趣的，请问大概啥时候能开源训练部分代码呢？

hi Ziang !
您好，如果我理解的没错的话SiamRPT模型最终的特征点的位置是通过两次offset相加得到的，请问如果只计算最后一次的offset的话实验结果会怎样呢，有做这样的实验吗？

hi Ziang !
很棒的工作！请问您训练部分的代码什么时候开源呢？

您好
我发现里面没有训练代码，想问下什么时候开源？我觉得这个工作很棒，想follow一下

作者你好，当我运行
cd siamreppoints/tools
python test.py
--snapshot ./snapshot/siamreppoints.model \ #model path
--dataset VOT2018 \ #dataset name
--config ./experiments/siamreppoints/config_vot2018_offline.yaml #config file
这里时，一直会报这个错，请问这个是哪里没有配置好吗还是其他原因，谢谢

Hi, I want to evaluate a tracker on the VOT-ST2020 dataset. The segmentation annotations are placed under the gorundtruth.txt but I could not understand the format and find an explanation for it. For example, the annotation format is as in below:

m242,164,118,115,29,1,2,1,110,45,73,61,56,69,47,76,41,81,36,83,34,85,33,86,32,86,32,87,30,88,29,90,28,90,27,92,26,92,26,93,25,93,25,94,23,95,22,97,20,98,20,98,19,100,18,100,18,100,17,101,17,102,15,103,15,104,14,104,14,104,13,105,13,105,10,109,9,109,9,109,9,110,8,111,7,112,6,112,6,111,6,112,5,113,5,113,4,115,1,117,1,117,1,117,1,115,3,115,3,115,3,115,3,115,3,115,3,115,3,115,3,115,3,115,2,116,3,115,3,115,2,116,3,115,3,115,3,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,1,117,1,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,116,2,115,3,16,3,96,3,16,3,14,10,72,3,16,3,13,18,65,3,16,5,10,20,13,26,25,3,16,9,3,24,9,30,24,3,16,38,6,31,24,3,17,38,3,34,23,3,17,76,22,3,17,76,22,3,16,77,22,3,15,78,22,3,14,79,22,3,15,78,22,3,14,79,22,4,12,80,22,4,11,82,21,5,9,84,20,99,19,100,17,102,15,104,12,107,10,8

What does m242 mean? And what do these points represent? I assume that they represent the segmentation pixel points but are they ordered like x,y,x,y...x,y pixel coordinates? I tried to create binary segmentation masks considering they are ordered x,y.. but the mask obtained by these points does not represent the object. Could you please explain the annotation format for VOTSTS2022?
I would be glad if you help me with this.

(rpt) C:\Users\jinda\OneDrive\Desktop\RPT-master>python eval.py --tracker_path ./results --dataset VOT2018 --tracker_prefix siamreppoints --num 1
loading VOT2018: 100%|██████████████████████████████████| 60/60 [00:00<00:00, 80.93it/s, zebrafish1]
eval ar: 100%|████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 4.52it/s]
eval eao: 0%| | 0/1 [00:00<?, ?it/s]
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
File "C:\Users\jinda\anaconda3\envs\rpt\lib\multiprocessing\pool.py", line 119, in worker
result = (True, func(*args, **kwds))
File "C:\Users\jinda\OneDrive\Desktop\RPT-master\toolkit\evaluation\eao_benchmark.py", line 47, in eval
eao = self._calculate_eao(tracker_name, self.tags)
File "C:\Users\jinda\OneDrive\Desktop\RPT-master\toolkit\evaluation\eao_benchmark.py", line 110, in _calculate_eao
max_len = max([len(x) for x in all_overlaps])
ValueError: max() arg is an empty sequence
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "eval.py", line 152, in
main()
File "eval.py", line 134, in main
trackers), desc='eval eao', total=len(trackers), ncols=100):
File "C:\Users\jinda\anaconda3\envs\rpt\lib\site-packages\tqdm\std.py", line 1176, in iter
for obj in iterable:
File "C:\Users\jinda\anaconda3\envs\rpt\lib\multiprocessing\pool.py", line 735, in next
raise value
ValueError: max() arg is an empty sequence

please help me to resolve this error
guide me where i am wrong

非常感谢

您好！请问您怎么获取VOT2020.json文件，方便共享么？我只能获取description.json文件 @zhanght021

Hi! Thank you for this amazing repository. I want to build my own backbone and integrate it into rpt, what are the necessary steps and can you please list all the files and directories where I will need to make changes. Thank you!

我用提供的raw results结果在vot2018测试，offline EAO是0.5020，online EAO只有0.4389

A great job! And could you please say when the complete code(online Classifier included) will be uploaded? Thank you.

Dear Zhanght,
Thank you for sharing your excellent codes.
I read the RPT paper. I'm curious about the how the representative points can be learned that indicates the object boundaries and semantically prominent regions.
According to my understanding from the paper, the training steps for the points are like below:

1. model estimates the positions of the points.
2. sudo box is obtained from points using min/max operations
3. overlap with ground-truth bbox is computed.
4. overlap-related loss is computed so model can learn from this loss.

Because there is only bbox overlap loss based on min/max operation, I guess the points could be moved to the bbox boundaries rather than object boundaries or semantically prominent positions.
I don't know how the points can be moved to the object boundaries and semantically prominent positions without other loss.
Could you please explain about it?
Thanks in advance.

I have questions about some hyperparameters in the config files.

experiments/siamreppoints/config_vot2018_offline.yaml:

TRACK:
    TYPE: 'SiamReppointsTracker'
    PENALTY_K: 0.09384699789214077
    WINDOW_INFLUENCE: 0.2870488747571366
    LR: 0.546949224973851

experiments/siamreppoints/config_vot2019_offline.yaml:

TRACK:
    TYPE: 'SiamReppointsTracker'
    PENALTY_K: 0.17380070743842363
    WINDOW_INFLUENCE: 0.43745716974756743
    LR: 0.4488453012609389

How did you determine those hyperparameters? Why do those values differ in the 2 files?

Hi，我按照您的论文中的描述复现了一下您的论文，但是我发现刚开始一阶段中是没有正样本的，导致第二阶段没了正样本来回归和分类，二阶段回归loss一直都是0，请问您是如何解决的。

Hello,
Can you upload the siamreppoint.model on Google Drive? I couldn't download the model from the provided website(Baidu website).

Hi. I downloaded your code and paper.
As you paper says: “Siamese-based feature extraction and matching”.How to do feature extraction and matching in paper? Can you explain it in detail?
Look forward to your replay. Thx.

在配置环境的最后一步：
python ./siamreppoints/setup.py build_ext --inplace
出现错误：error: command '/usr/local/cuda-9.0/bin/nvcc' failed with exit status 1
所有步骤都是按照建议进行的，网上查了很多资料也还是没有解决。
大家有知道如何解决的吗？

Hello, I'd like to use your RPT model on Got10k data, but for that purposes I need the GOT-10k.json file you created. I have no access to Baidu, could you please place it on google drive?
In addition, I want to test the model on my own data, so can I also ask you to send the script on creating GOT-10k.json file? Thank you!