Vladimir Iglovikov
Vladimir Iglovikov
Transform used by @ZFTurbo 1. Specify the distance from 4 sides in pixels or in percent. 2. Sample distance from `[0, specified_distance]` 3. Crop the corresponding part of the image/mask....
Similar to RandomErasing / CoarseDropout, etc, but instead of erasing the region replaces it with the grayscale version
Following https://arxiv.org/pdf/1708.04896.pdf and https://github.com/zhunzhong07/Random-Erasing - [x] Add an option for a random fill value - [ ] Add an option for object aware erasing
When we have instance segmentation we have a mask for every object => we can deduce the bounding box from the mask after the rotation. => more accurate bounding boxes
1. Specify the object as PNG with the empty background. 2. Specify background image. 3. Cut out a region in the background image that is larger than the object. 4....
There is a set of Pixel Level transforms that is used in the work `Benchmarking Neural Network Robustness to Common Corruptions and Perturbations` The authors also share the code =>...
There are WebP and Jpeg Compression transforms, but we can add heic, jpeg2000 and even video compression.
Example: We want to augment the color of T-shirts. => if we have a binary mask representing T-shirts, we can apply color transforms only to the region where `mask ==...
Similar to Kornia: https://kornia.readthedocs.io/en/latest/augmentation.module.html#transforms2d