手把手教你跑通6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints

项目地址:GitHub - j96w/6-PACK: "6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints" code repository

一、配置环境

   这里强烈建议安装python3.6.7版本,安装低于此版本的python有可能在多线程训练时造成
chunk = read(handle, remaining)的问题。

  • CUDA 9.0
    已经有其他版本CUDA的老铁们不要怕麻烦,最好还是再装个9.0然后多版本切换cuda。ubuntu下安装多版本cuda及版本切换教程_zeeq_的博客-CSDN博客_ubuntu切换cuda版本

  • conda python3.6.7环境
    conda create -n 6pack python==3.6.7
  • 安装torch0.4.1 torchvision0.2.1

    conda install pytorch=0.4.1 torchvision==0.2.1 cuda90 -c pytorch
  • 安装opencv 

    pip install opencv-contrib-python
  • 安装PIL/spicy/numpy/logging     from conda官网安装

    conda install -c conda-forge pillow
    conda install -c conda-forge scipy
    conda install -c conda-forge numpy
    conda install -c conda-forge r-logging

二、准备数据

6PACK需要的训练+测试数据很庞大(快200G了),建议提前一晚下载

  • 创建My_NOCS文件夹,将上述数据集数据解压其中(记得给Test解压出来的文件改个名为real_val)由于数据量很大,可以把My_NOCS文件创建在空间充足的硬盘里,然后添加软链接到6pack根目录。ubuntu创建软链接_卒迹的博客-CSDN博客_ubuntu软连接
  • 现在文件夹的形式为My_NOCS/data/real_train, My_NOCS/data/real_val, My_NOCS/data/train,My_NOCS/data/gts
  • 下载NOCS-REAL275-additional.zip并解压到My_NOCS
      由于nocs数据集里没有位姿真值,只有NOCS的真值(每个像素对应的点云在NOCS系下的位置),因此需要从训练数据里生成6D pose。NOCS的原版策略:随机采样mask内的像素,恢复出的真值效果一般。6PACK策略:选取临近mask中心的像素。additional.zip是6-pack官方已经生成好的pose和scale,model_pts是ShapeNetCore中的物体点云。
  • 现在文件夹下有My_NOCS/data_list,My_NOCS/data_pose,My_NOCS/model_pts,My_NOCS/model_scales
  • 在6pack根目录的dataset文件夹下运行命令,生成真值数据,时间较长,我用了6小时。
    python data_preprocess.py
  • 训练时采用数据增强策略,需要用到coco2017数据集,解压后把train2017放到My_NOCS下http://images.cocodataset.org/zips/train2017.zip

至此数据准备工作完成,即构建好了 My_NOCS文件夹。

三、训练

python train.py --category 1 --dataset_root '/home/robot413/My_NOCS' --workers 16

四、评估+benchmark

python eval.py
python benchmark.py

  • 2
    点赞
  • 5
    收藏
    觉得还不错? 一键收藏
  • 3
    评论
Color-based model-free tracking is a popular technique used in computer vision to track objects in video sequences. Despite its simplicity, it has demonstrated high accuracy and robustness in various applications, such as surveillance, sports analysis, and human-computer interaction. One of the key advantages of color-based model-free tracking is its real-time performance. Unlike model-based tracking, which requires complex training and computation, color-based tracking can be implemented using simple algorithms that can run in real-time on low-power devices. This makes it suitable for applications that require fast response time, such as robotics and autonomous systems. Another advantage of color-based tracking is its ability to handle occlusions and partial occlusions. Since color features are less sensitive to changes in lighting and viewing conditions, the tracker can still maintain its accuracy even when the object is partially hidden or obstructed by other objects in the scene. Critics of color-based tracking argue that it is not effective in complex scenes where the object of interest may have similar colors to the background or other objects in the scene. However, recent advancements in machine learning and deep learning have enabled the development of more sophisticated color-based tracking algorithms that can accurately detect and track objects even in challenging scenarios. In summary, color-based model-free tracking is a simple yet effective technique for tracking objects in video sequences. Its real-time performance, robustness, and ability to handle occlusions make it a popular choice for various applications. While it may not be suitable for all scenarios, advancements in machine learning are making it more effective in complex scenes.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 3
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值