add pose_estimation; add sample data and demo code

yqtl · Sep 18, 2016 · d94666c · d94666c
1 parent 419d595
commit d94666c
Show file tree

Hide file tree

Showing 100 changed files with 2,212 additions and 849 deletions.
diff --git a/README.md b/README.md
@@ -2,13 +2,55 @@
 
 
 
+## A Quick-Start: Matlab Demo
+Estimates 6D object poses on the sample scene data (in `data/sample`) with pre-computed object segmentation results from [Deep Learning FCN ROS Package](#deep-learning-fcn-ros-package).
+1. 
+
+
 ## Documentation
+* [6D Pose Estimation ROS Package](#6d-pose-estimation-ros-package)
 * [Realsense Standalone](#realsense-standalone)
 * [Realsense ROS Package](#realsense-ros-package)
 * [Deep Learning FCN ROS Package](#deep-learning-fcn-ros-package)
 * [FCN Training with Marvin](#fcn-training-with-marvin)
 * [Evaluation Code](#evaluation-code)
 
+## 6D Pose Estimation ROS Package
+A Matlab ROS Package for estimating 6D object poses by model-fitting with ICP on RGB-D object segmentation results.
+
+### Dependencies
+1. [Deep Learning FCN ROS Package](#deep-learning-fcn-ros-package) and all of its respective dependencies.
+2. Recommended: Matlab 2015b or later
+
+### Compilation
+1. Copy the ROS package `ros_packages/.../pose_estimation` into your catkin workspace source directory (e.g. `catkin_ws/src`)
+2. Follow the instructions on the top of `pose_estimation/src/make.m` to compile ROS custom messages for Matlab
+3. Compile a GPU CUDA kernel function in `pose_estimation/src`:
+```shell
+nvcc -ptx KNNSearch.cu
+```
+
+### Usage
+* Start `roscore`
+* To start the pose estimation service, run `pose_estimation/src/startService.m`. At each call (see service request format described in `pose_estimation/srv/EstimateObjectPose.srv`), the service:
+ * Calibrates the camera poses of the scene using calibration data
+ * Perform 3D background subtraction
+ * For each object in the scene, use model-fitting to estimate its 6D pose
+
+### Demo
+1. Install all dependencies and compile this package
+2. Start `roscore` in terminal
+3. Create a temporary directory to be used by marvin_convnet for reading RGB-D data and saving segmentation masks
+ * `mkdir /path/to/your/data/tmp`
+4. `rosrun marvin_convnet detect _read_directory:="/path/to/your/data/tmp"`
+5. Navigate to `pose_estimation/src`
+6. Edit file paths and options on the top of `demo.m`
+6. Open Matlab and run:
+```shell
+startService.m
+demo.m
+```
+
 ## Realsense Standalone
 
 A standalone C++ executable for streaming and capturing data (RGB-D frames and 3D point clouds) in real-time using [librealsense](https://github.com/IntelRealSense/librealsense). Tested on Ubuntu 14.04 and 16.04 with an Intel® RealSense™ F200 Camera.
@@ -63,7 +105,7 @@ See `ros-packages/realsense_camera`
  * Used for saving point clouds
 
 ### Compilation
-1. Copy the ROS package `ros_packages/realsense_camera` into your catkin workspace source directory (e.g. `catkin_ws/src`)
+1. Copy the ROS package `ros_packages/.../realsense_camera` into your catkin workspace source directory (e.g. `catkin_ws/src`)
 2. If necessary, configure `realsense_camera/CMakeLists.txt` according to your respective dependencies
 3. In your catkin workspace, compile the package with `catkin_make` 
 4. Source `devel/setup.sh`
@@ -105,26 +147,37 @@ sudo cp cuda/include/* /usr/local/cudnn/v5/include/
  * Used for saving images
 
 ### Compilation
-1. Copy the ROS package `ros_packages/marvin_fcn` into your catkin workspace source directory (e.g. `catkin_ws/src`)
+1. Copy the ROS package `ros_packages/.../marvin_convnet` into your catkin workspace source directory (e.g. `catkin_ws/src`)
 2. If necessary, configure `realsense_camera/CMakeLists.txt` according to your respective dependencies
 3. In your catkin workspace, compile the package with `catkin_make` 
 4. Source `devel/setup.sh`
 
-change location of where net is loaded
+### Usage
+* Navigate to `models/competition/` and run bash script `./download_weights.sh` to download our trained weights for object segmentation (trained on our [training dataset](http://www.cs.princeton.edu/~andyz/apc2016))
+* Edit `marvin_convnet/src/detect.cu`: Towards the top of the file, specify the filepath to the network architecture .json file and .marvin weights.
+* Create a folder called `tmp` in `apc-vision-toolbox/data` (e.g. `apc-vision-toolbox/data/tmp`). This where marvin_convnet will read/write RGB-D data. The format of the data in `tmp` follows the format of the scenes in our [datasets](http://www.cs.princeton.edu/~andyz/apc2016) and the format of the data saved by [Realsense Standalone](#realsense-standalone).
+* marvin_convnet offers two services: `save_images` and `detect`. The former retrieves RGB-D data from the [Realsense ROS Package](#realsense-ros-package) and writes to disk in the `tmp` folder, while the latter reads from disk in the `tmp` folder and feeds the RGB-D data forward through the FCN and saves the response images to disk 
+* To start the RGB-D data saving service, run: 
 
-make sure where data is going to be read and written
+```shell
+rosrun marvin_convnet save_images _write_directory:="/path/to/your/data/tmp" _camera_service_name:="/realsense_camera"
+```
 
-ros package to compute hha 
+* To start the FCN service, run:
 
-`rosrun marvin_convnet detect _service_mode:=1 _write_directory:="/home/andyz/apc/toolbox/data/tmp"`
+```shell
+rosrun marvin_convnet detect _read_directory:="/path/to/your/data/tmp" _service_name:="/marvin_convnet"
+```
 
-`rosrun marvin_convnet detect _service_mode:=2 _read_directory:="/home/andyz/apc/toolbox/data/tmp" _write_directory:="/home/andyz/apc/toolbox/data/tmp"`
+* Example ROS service call to do object segmentation for glue bottle and expo marker box (assuming the scene's RGB-D data is in the `tmp` folder):
 
-`rosservice call /marvin_convnet ["elmers_washable_no_run_school_glue","expo_dry_erase_board_eraser"] 0 0`
+```shell
+rosservice call /marvin_convnet ["elmers_washable_no_run_school_glue","expo_dry_erase_board_eraser"] 0 0
+```
 
 ## FCN Training with Marvin
 
-Code and models for training object segmentation using [FCNs (Fully Convolutional Networks)](https://arxiv.org/abs/1411.4038) with [Marvin](http://marvin.is/), a lightweight GPU-only neural network framework. Includes network architecture .json files in `convnet-training/models` and a Marvin data layer in `convnet-training/apc.hpp` that randomly samples images (RGB and HHA) from the segmentation training dataset [here](http://www.cs.princeton.edu/~andyz/apc2016).
+Code and models for training object segmentation using [FCNs (Fully Convolutional Networks)](https://arxiv.org/abs/1411.4038) with [Marvin](http://marvin.is/), a lightweight GPU-only neural network framework. Includes network architecture .json files in `convnet-training/models` and a Marvin data layer in `convnet-training/apc.hpp` that randomly samples RGB-D images (RGB and HHA) from our [segmentation training dataset](http://www.cs.princeton.edu/~andyz/apc2016).
 
 See `convnet-training`
 
@@ -146,19 +199,20 @@ sudo cp cuda/include/* /usr/local/cudnn/v5/include/
  * Used for reading images
 
 ### Setup Instructions
-1. Download segmentation training dataset from [here](http://www.cs.princeton.edu/~andyz/apc2016)
-2. Specify training dataset filepath in APCData layer of network architecture in `convnet-training/models/train_shelf_color.json`
-3. Navigate to `convnet-training/models/weights/` and run bash script `./download_weights.sh` to download VGG pre-trained weights on ImageNet (see [Marvin](http://marvin.is/) for more pre-trained weights)
+1. Download our [segmentation training dataset](http://www.cs.princeton.edu/~andyz/apc2016)
+2. Navigate to directory `convnet-training/`
+2. Specify training dataset filepath in APCData layer of network architecture in `models/train_shelf_color.json`
+3. Navigate to `models/weights/` and run bash script `./download_weights.sh` to download VGG pre-trained weights on ImageNet (see [Marvin](http://marvin.is/) for more pre-trained weights)
 4. Navigate to `convnet-training/` and run in terminal `./compile.sh` to compile Marvin.
 5. Run in terminal `./marvin train models/rgb-fcn/train_shelf_color.json models/weights/vgg16_imagenet_half.marvin` to train a segmentation model on RGB-D data with objects in the shelf (for objects in the tote, use network architecture `models/rgb-fcn/train_shelf_color.json`).
 
 ## Evaluation Code
-Code used to perform the experiments in the paper - tests the full vision system on the 'Shelf & Tote' benchmark dataset.
+Code used to perform the experiments in the paper; tests the full vision system on the 'Shelf & Tote' benchmark dataset.
 
 See `evaluation`
 
 ### Setup Instructions
-1. Download the full 'Shelf & Tote' benchmark dataset from [here](http://www.cs.princeton.edu/~andyz/apc2016) and extract its contents to `apc-vision-toolbox/data/benchmark` (e.g. `apc-vision-toolbox/data/benchmark/office`, `apc-vision-toolbox/data/benchmark/warehouse', etc.)
+1. Download our 'Shelf & Tote' benchmark dataset from [here](http://www.cs.princeton.edu/~andyz/apc2016) and extract its contents to `apc-vision-toolbox/data/benchmark` (e.g. `apc-vision-toolbox/data/benchmark/office`, `apc-vision-toolbox/data/benchmark/warehouse', etc.)
 2. In `evaluation/getError.m`, change the variable `benchmarkPath` to point to the filepath of your benchmark dataset directory
 3. We have provided our vision system's predictions in a saved Matlab .mat file `evaluation/predictions.mat`. To compute the accuracy of these predictions against the ground truth labels of the 'Shelf & Tote' benchmark dataset, run `evaluation/getError.m`
 

diff --git a/data/sample/calibration/shelf/cam.poses.A.txt b/data/sample/calibration/shelf/cam.poses.A.txt
@@ -0,0 +1,90 @@
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000000 to frame-000007
+ 9.06193058e-01	  3.42857864e-02	 -4.21471976e-01	  1.54523532e-01	
+-9.89394920e-02	  9.86232759e-01	 -1.32498762e-01	  4.56576146e-02	
+ 4.11126646e-01	  1.61769681e-01	  8.97109498e-01	  4.00954915e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000001 to frame-000007
+ 9.74033990e-01	  1.72623355e-02	 -2.25742770e-01	  8.42798845e-02	
+-5.44936063e-02	  9.85651131e-01	 -1.59756985e-01	  5.61639449e-02	
+ 2.19745838e-01	  1.67910271e-01	  9.60998391e-01	  1.65755026e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000002 to frame-000007
+ 9.99941171e-01	  9.65149581e-03	  4.95001832e-03	 -6.76798065e-04	
+-8.57582876e-03	  9.82882741e-01	 -1.84032532e-01	  6.82766391e-02	
+-6.64147679e-03	  1.83979255e-01	  9.82907689e-01	  8.37727195e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000003 to frame-000007
+ 9.67112011e-01	  6.05468410e-03	  2.54278781e-01	 -1.01311126e-01	
+ 3.73703670e-02	  9.85484988e-01	 -1.65598289e-01	  6.27228817e-02	
+-2.51590567e-01	  1.69654586e-01	  9.52848103e-01	  2.41726814e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000004 to frame-000007
+ 8.85187346e-01	  9.49438568e-03	  4.65137849e-01	 -1.78356751e-01	
+ 6.75713285e-02	  9.86566465e-01	 -1.48730377e-01	  5.92754124e-02	
+-4.60301507e-01	  1.63084230e-01	  8.72654603e-01	  5.42887834e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000005 to frame-000007
+ 9.08533859e-01	 -9.44479893e-02	  4.06996074e-01	 -1.42541407e-01	
+ 9.68736841e-02	  9.95188207e-01	  1.46942527e-02	  5.83083812e-03	
+-4.06425536e-01	  2.60769830e-02	  9.13311707e-01	  3.42312243e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000006 to frame-000007
+ 9.75567915e-01	 -4.42249756e-02	  2.15200824e-01	 -7.68780799e-02	
+ 4.49305673e-02	  9.98988808e-01	  1.61447263e-03	  7.44182428e-03	
+-2.15054615e-01	  8.09406742e-03	  9.76568481e-01	  1.14108454e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000007 to frame-000007
+ 1.00000000e+00	  0.00000000e+00	  0.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  1.00000000e+00	  0.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  0.00000000e+00	  1.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000008 to frame-000007
+ 9.77786599e-01	  5.05112051e-02	 -2.03425627e-01	  7.39230849e-02	
+-5.01212684e-02	  9.98718104e-01	  7.07162812e-03	 -5.35480146e-03	
+ 2.03522053e-01	  3.28140726e-03	  9.79064863e-01	  6.79063607e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000009 to frame-000007
+ 9.06949455e-01	  9.70696317e-02	 -4.09902638e-01	  1.50787138e-01	
+-9.18811102e-02	  9.95243091e-01	  3.23890512e-02	 -1.23167106e-02	
+ 4.11096762e-01	  8.28707713e-03	  9.11554045e-01	  3.04464373e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000010 to frame-000007
+ 9.04354087e-01	  1.57968695e-01	 -3.96471408e-01	  1.47415385e-01	
+-8.15376054e-02	  9.75816796e-01	  2.02813212e-01	 -6.94670376e-02	
+ 4.18921598e-01	 -1.51087628e-01	  8.95364297e-01	  3.55340376e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000011 to frame-000007
+ 9.76539943e-01	  7.69014782e-02	 -2.01136525e-01	  7.61511874e-02	
+-4.03811507e-02	  9.82885646e-01	  1.79736388e-01	 -6.45838775e-02	
+ 2.11516197e-01	 -1.67397638e-01	  9.62932463e-01	  1.10402609e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000012 to frame-000007
+ 9.99998641e-01	 -1.56633748e-03	 -5.15200283e-04	 -6.83942273e-05	
+ 1.63268064e-03	  9.84298747e-01	  1.76503003e-01	 -5.78220348e-02	
+ 2.30647723e-04	 -1.76503604e-01	  9.84299967e-01	  5.96666405e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000013 to frame-000007
+ 9.75575029e-01	 -8.57247925e-02	  2.02248911e-01	 -7.19359239e-02	
+ 5.06092233e-02	  9.83654260e-01	  1.72809152e-01	 -5.54876912e-02	
+-2.13757031e-01	 -1.58352634e-01	  9.63966999e-01	  1.39479082e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000014 to frame-000007
+ 8.99900305e-01	 -1.58189084e-01	  4.06393473e-01	 -1.49015967e-01	
+ 8.56453963e-02	  9.77850486e-01	  1.90979825e-01	 -5.60021794e-02	
+-4.27602978e-01	 -1.37057073e-01	  8.93516117e-01	  4.19620656e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
diff --git a/data/sample/calibration/shelf/cam.poses.B.txt b/data/sample/calibration/shelf/cam.poses.B.txt
@@ -0,0 +1,90 @@
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000000 to frame-000007
+ 9.05247257e-01	  2.89516540e-02	 -4.23897635e-01	  1.49310311e-01	
+-9.97035745e-02	  9.84292822e-01	 -1.45694330e-01	  5.84530536e-02	
+ 4.13021308e-01	  1.74153502e-01	  8.93914961e-01	  4.00667733e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000001 to frame-000007
+ 9.76027409e-01	  1.84703952e-02	 -2.16862495e-01	  7.64990842e-02	
+-5.62802670e-02	  9.83922320e-01	 -1.69497492e-01	  6.38484133e-02	
+ 2.10245164e-01	  1.77639277e-01	  9.61374671e-01	  1.52328476e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000002 to frame-000007
+ 9.99939471e-01	  1.01612775e-02	  4.21927314e-03	 -3.39140620e-03	
+-9.23715004e-03	  9.83659092e-01	 -1.79803963e-01	  6.77236939e-02	
+-5.97736435e-03	  1.79754106e-01	  9.83693414e-01	  7.54003432e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000003 to frame-000007
+ 9.73498541e-01	 -4.48275288e-03	  2.28649285e-01	 -8.80713433e-02	
+ 4.76295117e-02	  9.81858050e-01	 -1.83538003e-01	  7.15631809e-02	
+-2.23678386e-01	  1.89564432e-01	  9.56050891e-01	  2.05026384e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000004 to frame-000007
+ 8.96544327e-01	 -1.46288801e-02	  4.42712397e-01	 -1.66263642e-01	
+ 9.87479638e-02	  9.80903326e-01	 -1.67563434e-01	  7.03188472e-02	
+-4.31806797e-01	  1.93944994e-01	  8.80867884e-01	  4.99335242e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000005 to frame-000007
+ 9.05536125e-01	 -8.34244741e-02	  4.15986397e-01	 -1.53295248e-01	
+ 8.94276848e-02	  9.95980412e-01	  5.07022091e-03	  4.25375109e-03	
+-4.14737284e-01	  3.26094322e-02	  9.09356701e-01	  3.79101188e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000006 to frame-000007
+ 9.76479572e-01	 -4.14692150e-02	  2.11584378e-01	 -7.83460624e-02	
+ 4.43033703e-02	  9.98980503e-01	 -8.66982143e-03	  5.93579989e-03	
+-2.11009138e-01	  1.78398046e-02	  9.77321280e-01	  1.13443493e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000007 to frame-000007
+ 1.00000000e+00	  0.00000000e+00	  0.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  1.00000000e+00	  0.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  0.00000000e+00	  1.00000000e+00	  0.00000000e+00	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000008 to frame-000007
+ 9.78630296e-01	  5.52592281e-02	 -1.98063527e-01	  6.69496630e-02	
+-5.35004321e-02	  9.98466509e-01	  1.42244501e-02	 -4.07535827e-03	
+ 1.98545830e-01	 -3.32399352e-03	  9.80085968e-01	  6.43911432e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000009 to frame-000007
+ 9.14591958e-01	  1.00486098e-01	 -3.91693878e-01	  1.33637585e-01	
+-9.54076667e-02	  9.94908817e-01	  3.24626480e-02	 -3.41561345e-03	
+ 3.92961738e-01	  7.68052220e-03	  9.19522747e-01	  2.85597281e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000010 to frame-000007
+ 9.05477435e-01	  1.57156091e-01	 -3.94224020e-01	  1.41790888e-01	
+-7.92369450e-02	  9.75178771e-01	  2.06755584e-01	 -6.22180308e-02	
+ 4.16931794e-01	 -1.55975409e-01	  8.95454941e-01	  3.54998236e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000011 to frame-000007
+ 9.77589946e-01	  8.00082873e-02	 -1.94721781e-01	  6.67172086e-02	
+-4.35950010e-02	  9.81853397e-01	  1.84562680e-01	 -5.99440969e-02	
+ 2.05954787e-01	 -1.71937724e-01	  9.63337970e-01	  1.09906624e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000012 to frame-000007
+ 9.99960437e-01	 -2.84680346e-03	  8.42732460e-03	 -5.87763588e-03	
+ 1.34444403e-03	  9.84890233e-01	  1.73174543e-01	 -5.68454979e-02	
+-8.79298358e-03	 -1.73156362e-01	  9.84855095e-01	  4.41092430e-03	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000013 to frame-000007
+ 9.73023063e-01	 -8.22832744e-02	  2.15535571e-01	 -8.16399246e-02	
+ 4.55981708e-02	  9.84396366e-01	  1.69954706e-01	 -5.81909463e-02	
+-2.26156862e-01	 -1.55541820e-01	  9.61592333e-01	  1.35834602e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+
+# Camera-to-camera extrinsic matrix (camera pose) from frame-000014 to frame-000007
+ 9.01818422e-01	 -1.67823869e-01	  3.98194279e-01	 -1.44817903e-01	
+ 9.78829696e-02	  9.76885125e-01	  1.90037830e-01	 -6.06841190e-02	
+-4.20882951e-01	 -1.32403177e-01	  8.97400100e-01	  4.05301648e-02	
+ 0.00000000e+00	  0.00000000e+00	  0.00000000e+00	  1.00000000e+00	
+