Skip to content

Latest commit

 

History

History
20 lines (18 loc) · 3.74 KB

change_log.md

File metadata and controls

20 lines (18 loc) · 3.74 KB
Raw

History of PPLandet

In this file, we show the development of the PPLanedet.

History

  • [2023-02-09] : We updated some common models in pplanedet and released the convnext backbone. In the following days, we will focus on solving bugs in CLRNet. Version5 is coming soon. In version5, more bakcbones will be reproduced. We are doing some researches about bag-of-freebies in lane detection and arxiv paper is under preparing.
  • [2023-02-06] : We fixed some bugs in detect.py and added some data augmentation methods（e.g.GuassianBulr).
  • [2023-01-20] : We released the code of CLRNet. However, there still exists some bugs(NMS cuda code meets error) in CLRNet during inference. We will fix this bug soon.
  • [2023-01-16] 🔥 We released the version4. In v4, we reproduced the CondLane, a state-of-the-art lane detection method based on Keypoint. Meanwhile, we fixed some bugs in albumentations and made the albumentations be able to used in Keypoint based method. In order to facilitate the scientific research, we reproduced the RTFormer, a SOTA real-time semantic segmentation with Transformer.
  • [2023-01-20] : We released the code of CLRNet.
  • [2023-01-16] 🔥 We released the version4. In v4, we reproduced the CondLane, a state-of-the-art lane detection method based on Keypoint. Meanwhile, we fixed some bugs in albumentations and made the albumentations be able to used in Keypoint based method. In order to facilitate the scientific research, we reproduced the RTFormer, a SOTA real-time semantic segmentation with Transformer.
  • [2023-01-01] : We released the DeepLabV3+. Version 4 is coming soon. We will also open source the Colab demo of the PPLanedet.
  • [2022-12-29] : We reproduced RESA and pretrain weight is available. We also fixed some bugs in detect.py
  • [2022-12-23] : We reproduced the ERFNet and the weight is available. We also support the visualization of the segmentation results and code is shown in detect.py
  • [2022-12-19] : Versionv3 has been released. In this version, we reproduce the UFLD and the pretrain weight is available.
  • [2022-12-14] : We released versionv2. Compared with v1, v2 is achieved by Hook instead of being built upon Paddleseg. With v2, we can obtain a better SCNN with 95% accuracy on Tusimple dataset. It should be noticed that we only spent 30 epochs to achieve this result. Pretrain weight is available.
  • [2022-12-4] : We released the inference/demo code. You can directly test our model.
  • [2022-11-24] : We released the evaluation code and pretrain weight of the SCNN in Tusimple dataset. We also updated the Installation and training documentations of our project. In the following days, we will upload Inference/demo code and pretrain weight of SCNN in CULane dataset. Meanwhile, we will reproduce ERFNet.
  • [2022-11-22] We released the project code. We now only reproduce the SCNN with 93.70% accuracy in Tusimple dataset. Pretrain model will be updated in the following days. We will also release the eval and demo code in the following days.