delete_background_single_image.py: segmentation for a single image.delete_background_multimages.py:segmentation for a batch of images.test_trained_svm.py: this class trains a SVM model for the detection of the hand. You should not really care about this step, since it is already done.
Place RGBD depth masks in depth_images and their respective images into the 'images' directory. Then, run delete_background_multimages.py.
Results will be stored in the folders results, predilations and delations.
This is what happens to each image:
- First, the background is deleted exploiting depth information. (This step is stored into the
resultsfolder). - Then, SVM detects pixels belonging to the hand and tries to remove them. (This step is stored into the
predilationsfolder. - Dilation is applied for denoising the final image. (This step is stored into the
dilationsfolder).
| Original image (128x128) | Deleting the background and the holding hand | Binary image | Applying dilation |
|---|---|---|---|
 |
 |
 |
 |
-
Vincenzo Lomonaco and Davide Maltoni. "CORe50: a new Dataset and Benchmark for Continuous Object Recognition". Proceedings of the 1st Annual Conference on Robot Learning, PMLR 78:17-26, 2017.