tyyang123 / roam Goto Github PK

Code for "ROAM: Recurrently Optimizing Tracking Model [CVPR2020]"

License: MIT License

Python 100.00%

roam's Introduction

ROAM: Recurrently Optimizing Tracking Model

Introduction

This is the PyTorch implementation of our ROAM tracker published in CVPR, 2020. Detailed comparision results can be found in the author's webpage

Prerequisites

Python 3.5 or higher
PyTorch 1.4.0 or higher

Path setting

Set proper root_dir in config.py accordingly in order to proceed the following step. Make sure that you place the tracking data properly according to your path setting.

Training

Download the ILSRVC data from the official website and extract it to proper place according to the path in config.py. Pretrained vgg-16.mat file can be download from here
Then run the python3 make_vid_info.py in to build the meta data file for ILSVRC data.
Run:

python3 experiment.py \
  --mGPUs \
  --epochs 20 \
  --bs [BATCH_SIZE] \
  --nw [NUM_WORKERS] \
  --lr_mi 1e-6 \
  --lr_mo 1e-3

to train the model. Note we train our model on a 4-GPUs machine with BATCH_SIZE=16

Tracking Demo

After training, you can run python3 demo.py to test our tracker.

Citing ROAM

If you find the code is helpful, please cite

@inproceedings{Yang2020cvpr,
	author = {Yang, Tianyu and Xu, Pengfei and Hu, Runbo and Chai, Hua and Chan, Antoni B},
	booktitle = {CVPR},
	title = {{ROAM: Recurrently Optimizing Tracking Model}},
	year = {2020}
}

roam's People

Contributors

Stargazers

Watchers

Forkers

deepbehavier xieyunjiao xrosliang maliangzhibi cv-ip qinhuaping sundarammuthu le-wei fasladodo sun1992-max minyuedayuan

roam's Issues

Hi, how to get the vgg-16 model for testing?

FileNotFoundError: [Errno 2] No such file or directory: '/Data/Pre-trained/vgg-16.mat'.

I tried one model from matconvnet, but it shows me the following errors:
File "demo.py", line 63, in
tracker.load_models(models[-1])
IndexError: list index out of range

So, would you please provide the link for the vgg-16 you used in this code? Thanks.

I have get the vgg-16 model for testing and but it didn't work

Hi, I follow the introduction of README.md to test the model but it still has a bug:
Traceback (most recent call last):
File "D:/ROAM-master/ROAM-master/demo.py", line 63, in
tracker.load_models(models[-1])
IndexError: list index out of range
Could you help to solve this problem?

How can i get the trained models in demo.py?

Seems that there needs a well-trained model for demo.py.
For comparison, could you kindly release this model?

What's the difference between θcf and θ reg?

Hi,thank you for your contribution!I read your article recently.
1.Except the initialization weights θ and the branches of the application are different.What's the difference between correlation filter θcf and bounding box regression filter θreg ?
2.如果θreg和 θcf都用双线性插值重新调整了目标特征图的大小，这跟摘要里提到的the resizable convolutional filters什么关系呀？文章里很多名词都用了resizable 修饰了，他们的主要根源都是因为用了卷积滤波器的双线性插值吗？抱歉我看文章看的有点晕了
3.根据这一段的描述"The tracking model contains two branches where the response generation branch determines the presence of target by predicting a confidence score map and the bounding box regression branch estimates the precise box of the target by regressing coordinate shifts from the box anchors mounted on the sliding-window positions. "
和从文章中的Figure1中看，预测目标存在的 the response generation branch 和the bounding box regression branch是两个平行的分支，这两个分支在元学习框架中的交界点一起更新了损失
。响应生成分支用来确定帧中目标是否存在的话，这个分支的结果有什么用呀？两个分支是各用各的结果吗，各自更新？
4.当LSTM模型更新的时候，文章中描述到将更新了的θ模型用到未来的随机帧数上测试模型的性能。。。这种随机帧的条件下，如果目标不存在于当前帧中的条件下，即使在未来帧更新了相应的模型θcf and θ reg，此时响应生成分支存在的意义是什么？
5.边框回归分支是从Anchor转换得到精确的矩形框，然后用这些转换来的框做边框回归的输入，进而预测未来帧中的目标边框，文章说由于这个框是resizable的就只设置了一个Anchor，是直接把VGG-16传过来的特征图的中心点作为Anchor吗？
6."....... where F is the feature map input......." 从VGG-16计算得来的特征映射图是如何计算的呢？这个特征映射图作为响应生成分支和回归分支的输入，请问这个特征大小是怎么设定的？

Where can I get pretrained model?

Can you share the pretrained model?

请问可以提供下vgg-16.mat的预训练模型吗？

抱歉，我又来打扰了~。。我找到啦