Python png2klg converter tools

I'm fucked with C++, so I write this tool with python.

I'm working on ElasticFusion and Kintinuous and want to make it work on my own data.

Requirements:

opencv
numpy
zlib
click
tqdm

Just do this:

pip install opencv-python numpy click tqdm

Maybe you will get many errors, and you need to install zlib, deal with it.

If you just want to use this tool, and the files are organized as below, use the following command:

||--png2klg
  ||--rgb_images
  ||--depth_images
  ||timestamps.txt
  ||main.py

python main.py

The number of color images and depth images need to be the same. You can use python main.py --help to get help. For example, you can provide your own image folder path instead of copying the files to this project. The timestamps.txt file is not a must, because refer to mp3guy/ElasticFusion#18, the author says "Unit-less, just make it an increasing count (so frame one has timestamp 1, etc...)."

python main.py --depth_path=test/depth --rgb_path=test/rgb

If you want to make it work simultaneously with realsense, refer to mp3guy/ElasticFusion#164 failed with realsense camera L515)

(New: I succeeded with realsense D435i. It works simultaneously with ElasticFusion and the result seems fine.)

About Realsense(I believe you have already made the camera work)(My camera is L515):

You can refer to https://github.com/IntelRealSense/librealsense/blob/master/wrappers/python/examples/align-depth2color.py to get aligned color images and depth images. I made some changes, you can use command python get_camera.py to get aligned images.

This command will make directoriesrgb_images and depth_images to save images and timestamps.txt to save timestamps.

I have upload some images in this folder, you can test it with command python main.py and you will get data.klg.

You can also directly use the data.klg(only 50 frames) for test.

The most import thing when saving depth images is:

depth_image = cv2.resize((depth_image * depth_scale).astype(np.uint16), (640, 480)). To get depth image with proper scale and dtype, I don't know why this, but it works(not really...).

Update:

I have tested with the code above, but seems not work fine. I really don't know what 'depth scale' this algorithm uses.

You can refer to carlosbeltran/klgparser#1 (My issue) to get more details.

About Ros(My lab is using ros, so I also want to get .klg file from .bag file):

I didn't write the bag2klg code. You can use rosbag(just pip install rosbag) to parse images from .bag file. Remember to sync the timestamp between color images and depth images.

Then use this tool.

Ros simultaneously with Elasticfusion?(don't know)

How to parse images from .klg file?

Please refer to https://github.com/carlosbeltran/klgparser, I learned the klg file protocol and code from this repo.