- Add more data source apis for open source datasets, involving classification, detection, segmentation and keypoints tasks. And part of the data source apis support automatic download. For more information, please refer to [data_hub](https://github.com/alibaba/EasyCV/blob/master/docs/source/data_hub.md) ([#206](https://github.com/alibaba/EasyCV/pull/206) [#229](https://github.com/alibaba/EasyCV/pull/229))
- Release YOLOX-PAI which achieves SOTA results within 40~50 mAP (less than 1ms) ([#154](https://github.com/alibaba/EasyCV/pull/154) [#172](https://github.com/alibaba/EasyCV/pull/172) [#174](https://github.com/alibaba/EasyCV/pull/174) )
- Releases imagenet1k, imagenet22k, coco, lvis, voc2012 data with BaiduDisk to accelerate downloading ([#145](https://github.com/alibaba/EasyCV/pull/145) )
- Add detection predictor which support model inference without exporting models([#158](https://github.com/alibaba/EasyCV/pull/158) )
- Add VitDet support for faster-rcnn ([#155](https://github.com/alibaba/EasyCV/pull/155) )
- Release YOLOX-PAI which achieves SOTA results within 40~50 mAP (less than 1ms) ([#154](https://github.com/alibaba/EasyCV/pull/154) [#172](https://github.com/alibaba/EasyCV/pull/172) [#174](https://github.com/alibaba/EasyCV/pull/174) )
- Support DINO algo ([#144](https://github.com/alibaba/EasyCV/pull/144))
- Self-Supervised support ConvMAE algorithm (([#101](https://github.com/alibaba/EasyCV/pull/101)) ([#121](https://github.com/alibaba/EasyCV/pull/121)))
- Classification support EfficientFormer algorithm ([#128](https://github.com/alibaba/EasyCV/pull/128))
- Detection support FCOS、DETR、DAB-DETR and DN-DETR algorithm (([#100](https://github.com/alibaba/EasyCV/pull/100)) ([#104](https://github.com/alibaba/EasyCV/pull/104)) ([#119](https://github.com/alibaba/EasyCV/pull/119)))
- Segmentation support UperNet algorithm ([#118](https://github.com/alibaba/EasyCV/pull/118))
- Support use torchacc to speed up training ([#105](https://github.com/alibaba/EasyCV/pull/105))
- Add **semantic segmentation** modules, support FCN algorithm ([#71](https://github.com/alibaba/EasyCV/pull/71))
- Expand classification model zoo ([#55](https://github.com/alibaba/EasyCV/pull/55))
- Support export model with **[blade](https://help.aliyun.com/document_detail/205134.html)** for yolox ([#66](https://github.com/alibaba/EasyCV/pull/66))
- Support **ViTDet** algorithm ([#35](https://github.com/alibaba/EasyCV/pull/35))
- Add sailfish for extensible fully sharded data parallel training ([#97](https://github.com/alibaba/EasyCV/pull/97))
- Support run with [mmdetection](https://github.com/open-mmlab/mmdetection) models ([#25](https://github.com/alibaba/EasyCV/pull/25))
## New Features
- Set multiprocess env for speedup ([#77](https://github.com/alibaba/EasyCV/pull/77))
- Add data hub, summarized various datasets in different fields ([#70](https://github.com/alibaba/EasyCV/pull/70))
## Bug Fixes
- Fix the inaccurate accuracy caused by missing the `groundtruth_is_crowd` field in CocoMaskEvaluator ([#61](https://github.com/alibaba/EasyCV/pull/61))
- Unified the usage of `pretrained` parameter and fix load bugs（([#79](https://github.com/alibaba/EasyCV/pull/79)) ([#85](https://github.com/alibaba/EasyCV/pull/85)) ([#95](https://github.com/alibaba/EasyCV/pull/95))
## Improvements
- Update MAE pretrained models and benchmark ([#50](https://github.com/alibaba/EasyCV/pull/50))
- Add detection benchmark for SwAV and MoCo-v2 ([#58](https://github.com/alibaba/EasyCV/pull/58))
- Add moby swin-tiny pretrained model and benchmark ([#72](https://github.com/alibaba/EasyCV/pull/72))
- Update prepare_data.md, add more details ([#69](https://github.com/alibaba/EasyCV/pull/69))
- Optimize quantize code and support to export MNN model ([#44](https://github.com/alibaba/EasyCV/pull/44))
EasyCV provides state-of-the-art algorithms in self-supervised learning based on contrastive learning such as SimCLR, MoCO V2, Swav, DINO and also MAE based on masked image modeling. We also provides standard benchmark tools for ssl model evaluation.
2. Vision Transformers
EasyCV aims to provide plenty vision transformer models trained either using supervised learning or self-supervised learning, such as ViT, Swin-Transformer and XCit. More models will be added in the future.
3. Functionality & Extensibility
In addition to SSL, EasyCV also support image classification, object detection, metric learning, and more area will be supported in the future. Although convering different area, EasyCV decompose the framework into different componets such as dataset, model, running hook, making it easy to add new compoenets and combining it with existing modules.
EasyCV provide simple and comprehensive interface for inference. Additionaly, all models are supported on PAI-EAS, which can be easily deployed as online service and support automatic scaling and service moniting.
3. Efficiency
EasyCV support multi-gpu and multi worker training. EasyCV use DALI to accelerate data io and preprocessing process, and use fp16 to accelerate training process. For inference optimization, EasyCV export model using jit script, which can be optimized by PAI-Blade.
