search

CVLAB-Unibo / Real-time-self-adaptive-deep-stereo

Code for "Real-time self-adaptive deep stereo" - CVPR 2019 (ORAL)
Apache License 2.0
420 stars 73 forks source link

ZED disparity quality issue #48

Closed yaobaishen closed 4 years ago

yaobaishen commented 4 years ago

Thanks for providing the Live_Adaptation_Demo.py and luckily I have a first generation ZED camera, not the ZED_mini. After some simple modifications of the grabber.py, I can run the demo online with ZED images (720P@30fps), but the disparity looks not good and is very unstable. Here are the steps:

  1. launch the demo indoor;
  2. move the ZED camera around my office to gather enough images for the model adapting;
  3. wait 3 to 5 minutes, so there are about 5000~8000 images which have been send to the network;
  4. the disparity is still not good. So, do I MUST use the ZED_mini camera? I think the baseline difference is not very big. Or is there some extra work I need to do? Thanks.
yaobaishen commented 4 years ago

here is the command line:

python3  Live_Adaptation_Demo.py  --weights ../pretrained_nets/MADNet/synthetic/weights.ckpt
AlessioTonioni commented 4 years ago

run python 3 Live_Adaptation_Demo.py -h to check all the available command line options and their default value.

In particular the --mode flag sets the adaptation strategy to use and defaults a NONE therefore if you don't set it properly the code is not running in adaptation mode but only inference.

Try to add --mode FULL or --mode MAD

yaobaishen commented 4 years ago

I have a try the

--mode FULL

option and the disparity has noticeable improvement, I need some time to do the evaluation and will update here, cheers ~