Icevision

Added
- `iou_thresholds` parameter to `COCOMetric`
- `SimpleConfusionMatrix` Metric
- Negative samples support for mmdetection object detection models

Changed
- **Breaking:** Albumentations `aug_tfms` defaults.
- rotate_limit changed from 45 to 15
- rgb_shift_limit changed from 20 to 10
- VOC parser uses image sizes from annotation file instead of image
- bumps fastai to latest version (<2.4)

The following PRs have been merged since the last version.

dnth
- [Minor fix for clarity in mmdet utils.py](https://github.com/airctic/icevision/issues/1061) (#1061)
- [Swin Transformer notebook](https://github.com/airctic/icevision/issues/1059) (#1059)
- [Closes bug in 1057](https://github.com/airctic/icevision/issues/1058) (#1058)
- [CentripetalNet Support](https://github.com/airctic/icevision/issues/1050) (#1050)
- [Add YOLACT support from mmdet](https://github.com/airctic/icevision/issues/1046) (#1046)
- [Add more Swin backbones for RetinaNet and VFNet](https://github.com/airctic/icevision/issues/1042) (#1042)
- [Fix Swin backbone issue with single stage models](https://github.com/airctic/icevision/issues/1039) (#1039)
- [Add SABL (Side-Aware Boundary Localization for More Precise Object Detection)](https://github.com/airctic/icevision/issues/1038) (#1038)
- [Add Feature Selective Anchor-Free Module for Single-Shot Object Detection (FSAF)](https://github.com/airctic/icevision/issues/1037) (#1037)
- [Add Swin Transformer backbone to all mmdet models](https://github.com/airctic/icevision/issues/1035) (#1035)
- [Support custom config for two stage detectors](https://github.com/airctic/icevision/issues/1034) (#1034)
- [Add Swin Transformer backbone for Mask RCNN](https://github.com/airctic/icevision/issues/1033) (#1033)
- [Add Deformable DETR model from mmdet](https://github.com/airctic/icevision/issues/1032) (#1032)
- [Add YOLOF mmdet model](https://github.com/airctic/icevision/issues/1030) (#1030)

FraPochetti
- [removing fiftyone as a hard dependency](https://github.com/airctic/icevision/issues/1055) (#1055)
- [fix progressive resizing nb + addings nbs to docs](https://github.com/airctic/icevision/issues/1023) (#1023)
- [fixing plot_top_losses and semantic seg](https://github.com/airctic/icevision/issues/1019) (#1019)
- [fixing the OCHuman notebook](https://github.com/airctic/icevision/issues/1008) (#1008)
- [fixing wrong colab badge in SAHI notebook](https://github.com/airctic/icevision/issues/986) (#986)
- [SAHI inference integration](https://github.com/airctic/icevision/issues/984) (#984)

ai-fast-track
- [Update FiftyOne NB and add it to documentation](https://github.com/airctic/icevision/issues/1052) (#1052)
- [Added YOLOX backbones](https://github.com/airctic/icevision/issues/1010) (#1010)
- [SSD update](https://github.com/airctic/icevision/issues/993) (#993)
- [Added training to the custom parser notebook](https://github.com/airctic/icevision/issues/989) (#989)
- [updated CHANGELOG.md to the last PR](https://github.com/airctic/icevision/issues/974) (#974)

potipot
- [fix efficientdet metrics failing test](https://github.com/airctic/icevision/issues/1022) (#1022)
- [Fix wandb_efficientdet notebook](https://github.com/airctic/icevision/issues/1021) (#1021)
- [Fix opencv colab issue](https://github.com/airctic/icevision/issues/1020) (#1020)
- [fixing the negative_samples notebook](https://github.com/airctic/icevision/issues/1013) (#1013)
- [Update installation docs](https://github.com/airctic/icevision/issues/995) (#995)

strickvl
- [Fix docstring typo](https://github.com/airctic/icevision/issues/1028) (#1028)
- [Fix broken link](https://github.com/airctic/icevision/issues/1027) (#1027)
- [Fix bullet list formatting error](https://github.com/airctic/icevision/issues/985) (#985)

matt-deboer
- [Properly restore transformed masks after unload](https://github.com/airctic/icevision/issues/981) (#981)
- [fix for 978](https://github.com/airctic/icevision/issues/980) (#980)
- [fix for 978: store transform results in detection.masks](https://github.com/airctic/icevision/issues/979) (#979)

fstroth
- [Remove redundant model creation](https://github.com/airctic/icevision/issues/1036) (#1036)
- [(Fix) Fixed the notebook and the draw records function as well a the …](https://github.com/airctic/icevision/issues/1018) (#1018)

hectorLop
- [refactor: Fixed the bug to add neck modules properly](https://github.com/airctic/icevision/issues/1029) (#1029)
- [Implementation of DETR using mmdetection](https://github.com/airctic/icevision/issues/1026) (#1026)

2649
- [(feat) Added utils to convert Records and Predictions to fiftyone](https://github.com/airctic/icevision/issues/1031) (#1031)

fcakyon
- [compatibility for latest sahi updates](https://github.com/airctic/icevision/issues/1015) (#1015)

gablanouette
- [Fix checkpoint loading when model_name contains models or additional components](https://github.com/airctic/icevision/issues/1005) (#1005)

Anjum48
- [Issue 987 - Add detection_threshold arg for all Lightning adapters](https://github.com/airctic/icevision/issues/1004) (#1004)

**Thank you to all contributers: dnth, FraPochetti, ai-fast-track, potipot, strickvl, matt-deboer, fstroth, hectorLop, 2649, fcakyon, gablanouette, Anjum48**

[comment]: (Version_end)



[comment]: (Version_start)

The following PRs have been merged since the last version.

ai-fast-track
- [Updating mmcv installation to torch 1.10.0](https://github.com/airctic/icevision/issues/972) (#972)
- [Upgrade to torch 1.10 and torchvision 0.11](https://github.com/airctic/icevision/issues/970) (#970)
- [Pass both map_location, and logger to downstream methods](https://github.com/airctic/icevision/issues/968) (#968)
- [Bumped torch and torchision versions](https://github.com/airctic/icevision/issues/961) (#961)
- [Update CHANGELOG.md for Release 0.11.0](https://github.com/airctic/icevision/issues/959) (#959)
- [Adding an installation script for cuda and cpu](https://github.com/airctic/icevision/issues/956) (#956)
- [fixed yaml issue in doc generation CI/CD](https://github.com/airctic/icevision/issues/952) (#952)
- [Upgrade mk-docs-build.yml in the CI/CD](https://github.com/airctic/icevision/issues/951) (#951)
- [Update mmcv to 1.3.14 and mmdet to 2.17.0 in CI/CD](https://github.com/airctic/icevision/issues/949) (#949)
- [Update notebooks installation](https://github.com/airctic/icevision/issues/940) (#940)
- [Fix Colab script](https://github.com/airctic/icevision/issues/938) (#938)
- [Fixed Colab installation script](https://github.com/airctic/icevision/issues/937) (#937)
- [Update installation to torch 1.9 and dependencies](https://github.com/airctic/icevision/issues/935) (#935)
- [Inference - automatically recreate model trained with COCO](https://github.com/airctic/icevision/issues/929) (#929)
- [Simplify save and load model checkpoints](https://github.com/airctic/icevision/issues/924) (#924)
- [Update installation to torch 1.9 + dependencies](https://github.com/airctic/icevision/issues/919) (#919)
- [Added MMDetection VFNet Support.](https://github.com/airctic/icevision/issues/906) (#906)
- [Make MMDetection config object accessible to users](https://github.com/airctic/icevision/issues/904) (#904)
- [Adding progressive resizing support](https://github.com/airctic/icevision/issues/902) (#902)
- [Fix mmdet weights path issue](https://github.com/airctic/icevision/issues/900) (#900)
- [add docker-compose instructions](https://github.com/airctic/icevision/issues/898) (#898)
- [Added script for icevision inference installation](https://github.com/airctic/icevision/issues/893) (#893)
- [Added kwargs and label_border_color to end2end_detect()](https://github.com/airctic/icevision/issues/891) (#891)
- [Fix icevision installation in Colab](https://github.com/airctic/icevision/issues/887) (#887)
- [added kwargs to the EfficientDet model() method](https://github.com/airctic/icevision/issues/883) (#883)

fstroth
- [(WIP) Fix masks for instance segmentation](https://github.com/airctic/icevision/issues/967) (#967)
- [(Refactor) Removed the coco function.](https://github.com/airctic/icevision/issues/964) (#964)
- [(Feature) init coco and via parser with a dict instead of the filepath](https://github.com/airctic/icevision/issues/963) (#963)
- [(Feature) Added way to output metrics for pytorchlightning during training](https://github.com/airctic/icevision/issues/960) (#960)
- [Fix for CHANGLOG.md update script.](https://github.com/airctic/icevision/issues/958) (#958)
- [Script for automatically updating CHANGELOG.md](https://github.com/airctic/icevision/issues/957) (#957)
- [(Update) Updated code to run with albumentations version 1.0.3.](https://github.com/airctic/icevision/issues/927) (#927)
- [Radiographic images](https://github.com/airctic/icevision/issues/912) (#912)

potipot
- [Fix show pred](https://github.com/airctic/icevision/issues/930) (#930)
- [Fix inference on rectangular efficientdet input](https://github.com/airctic/icevision/issues/910) (#910)

FraPochetti
- [adding docker support](https://github.com/airctic/icevision/issues/895) (#895)
- [Colab Install Script: fixing link to icevision master](https://github.com/airctic/icevision/issues/888) (#888)

jaeeolma
- [Empty mask fix](https://github.com/airctic/icevision/issues/933) (#933)

bogdan-evtushenko
- [Add support for yolox from mmdetection.](https://github.com/airctic/icevision/issues/932) (#932)

drscotthawley
- [casting both caption parts as str](https://github.com/airctic/icevision/issues/922) (#922)

lgvaz
- [Unet3](https://github.com/airctic/icevision/issues/907) (#907)

nicjac
- [Fixed PIL size bug in ImageRecordComponent (889)](https://github.com/airctic/icevision/issues/894) (#894)

**Thank you to all contributers: ai-fast-track, fstroth, potipot, FraPochetti, jaeeolma, bogdan-evtushenko, drscotthawley, lgvaz, nicjac**

[comment]: (Version_end)

Main dependencies updated
- torch 1.9.0
- tochvision 0.10
- mmdet 2.16.0
- mmcv 1.3.14
- fastai 2.5.2
- pytorch-lightning 1.4.8

Added
- Low level parsing workflow with `RecordCollection`
- Semantic segmentation support with fastai


Changed
- **Breaking:** Refactored mask components workflow
- **Breaking:** Due to the new mask components refactor, autofix doesn't work for mask components anymore.

Added
- `end2end_detect()`: Run Object Detection inference (only `bboxes`) on a single image, and return predicted boxes corresponding to original image size
-**Breaking:** BaseLabelsRecordComponent `as_dict()` now returns both `labels` `and labels_ids`. `labels` are now strings instead of integers.

Changed
- **Breaking:** On `tfms.A.aug_tfms` parameter `size` and `presize` changed from order (height, width) to (width, height)
- Added `RecordCollection`
- **Breaking:** Changed how the *resnet* (not-fpn) backbone cut is done for torchvision models. Previous *resnet torchvision* trained models will have trouble loading weights.