Vladimir Iglovikov
Vladimir Iglovikov
https://www.linkedin.com/feed/update/urn:li:activity:7259354845051002882 
People are not aware of: - clip - save_applied_params
Use case: tracking. We may have object moving, and to simulate occlusions, we cut it out. But only this object, not others.
To question on why the number of keypoints or bounding boxes increased https://github.com/albumentations-team/albumentations/issues/2055 Need to add clear explanation with example to docs about the behavior of transforms with Reflection padding
Was told that noone is using Albumentation in the audio community, although all transforms from torchaudio exist in Albumentations, although may have different names. Need document / blog post.
It would be nice to reverse engineer this transformation and implement in albumentations: https://www.cambridgeincolour.com/tutorials/levels.htm
1. Fixed broken link to the list of transforms in torchvison. 2. Extended section about video image augmentations with an example from Albumentations.
### Describe the feature and motivation Currently cv2.warpAffine supports only 1, 3 and 4 channels images, while for problems of medical or satellite imagery we may have any. Would be...
### Describe the feature and motivation Currently cv2.warpAffine does not support `cv2.INTER_NEAREST_EXACT`, `cv2.INTER_LINEAR_EXACT` interpolation modes. But for problems of image segmentations, where pixel level accuracy is important for prediction values...