Prokofiev Kirill

Update XPU install

1

### Summary ### How to test ### Checklist - [ ] I have added unit tests to cover my changes.​ - [ ] I have added integration tests to cover...

### Summary Motivation: 1. Early stopping from lightning doesn't have a warmup parameter and it leads to underfitting when training on small Geti target datasets. I added two thresholds to...

Instance Segmentation Model refactoring

7

### Summary - [x] Remove code duplication for torchvision MascRCNN - [x] Add factory template for all models - [x] Move loss out of the heads - [x] Remove mmdict...

Move transforms from 3d object detection dataset

1

### Summary Move - [x] Move affine transforms including resize from the dataset to torchvision transforms - [x] Remove kitti utils - [x] PIL -> OpenCV, align image ratio -...

Fix export for problematic models

4

### Summary resolves https://github.com/open-edge-platform/training_extensions/issues/5073 ### How to test ### Checklist - [ ] The PR title and description are clear and descriptive - [ ] I have manually tested the...

Add DEIMV2 Object Detection Model

9

### Summary resolves [#5015 ](https://github.com/open-edge-platform/training_extensions/issues/5015) - [x] Add DinoV3 and VIT tiny as a backbones for detection, primarily for DeimV2 model - [x] Add DEIMV2 model (OTXModel, Encoder, Decoder), e2e...

Current values for MaskRCNN looks too strict, making training on difficult datasets suboptimal Related reported issue: https://github.com/open-edge-platform/geti/issues/1603 Consider changing them and check other models as well

[WIP] Upgrade Augmentations Pipeline in OTX

1

### Summary resolves https://github.com/open-edge-platform/training_extensions/issues/5020 - [ ] Provide batch augmentations support per model per task - [ ] Change augmentation pipeline to hybrid mode (batch augmentation + torchvision.v2). Operate only...

Currently, models are trained with unusual normalization values. Correcting these values could enhance the model’s performance and robustness. For YOLOX, we currently do not use any normalization. For MaskRCNN-EfficientB2B, we...

### 📄 Description **Current Problem:** 1. **Mixed augmentation pipelines**: OTX currently uses a combination of `torchvision` and OpenMMLab augmentations (some self-implemented in the repo). Later in the pipeline, operations rely...