将会在 B 站开启一场直播,通过实际案例,教你从数据收集、数据半自动标注、模型训练及调优到模型手机部署,实现一个目标检计数方案,并深入介绍如何应用 VisualDL 可视化分析工具对训练参数、网络结构等进行分析,从而指导开发者快速理解训练过程、进行算法优化。由此项目,可延伸泛化到「工业零件检测计数」『人流量统计』等领域应用。
**This is a PRE-RELEASE version of VisualDL 2.0. Please DO NOT use it in production environment.**
## Improvements
-**backend:** improve performance of plugin scalar and image by uploading logs incrementally
## Features
-**frontend:** completely newly-designed web user interface (for more information, please refer to [Frontend Readme](https://github.com/PaddlePaddle/VisualDL/blob/develop/frontend/README.md))
## BREAKING CHANGES
-`Histogram` is not available in this version.
- Audio and text samples are not available in this version.
- Plugin image displays 10pcs sampled images in this version.
# RELEASE 1.3.0
## New Features
- Support show paddle model,user can use **paddle.fluid.io.save_inference_model()** to save their model to <SAVE_MODEL_PATH>, and then use **visualdl --model_pb <SAVE_MODEL_PATH>** to load the model
- Change the dependent version of protobuf to 3.1.0
# RELEASE 1.2.1
## New Features
- Add language Selection options. Users can use -L or --language option to select the language show in the webpage(1.2.1 only support {zh, en}, and default is en)
# RELEASE 1.2.0
## New design
- Cleaner and more organized interface
- Show / select runs in global bar instead of individual tab
- Merge Scalar and Histogram into Metrics
- Merge Image, Audio and Text into Samples
- New Config design, add filter types in config
- New tags bar design instead of expand panel, easier to filter by tags
- New search bar integrated on tags bar
- Improve performance by showing less duplicate charts with tabs bar design
# RELEASE 1.1.0
## New Features
- Add new `High Dimensional / embeddings` recording visualization. Users can visualize data embeddings by projecting high-dimensional data into 2D / 3D
- Add Caffe2 Demo and tutorial
- Update `vdl_create_log`, add embeddings data and ONNX model in scratch_log
- Windows support
## Bug Fixes and improvements
- Fix issues where different Python libraries conflicted in Mac
- Update Visual DL new logo
# RELEASE 1.0.0
## New Features
- Improve the Graph feature to include interactive response. Users can now click on a node to inspect its details such as parameters, input, output, dimensions, etc. They can also zoom in/out and move around if the graph is too large, restore to its original size and download it as PNG. - Add new `High Dimensional / embeddings` recording visualization. Users can visualize data embeddings by projecting high-dimensional data into 2D / 3D
- Add new `Audio` recording visualization. Users can recording audio and inspect to help fine tuning the audio training. - Add Caffe2 Demo and tutorial
- Add new `Text` recording visualization. Users can record strings during training and inspect later. - Update `vdl_create_log`, add embeddings data and ONNX model in scratch_log
- User can manually `save` to sync the recordings to the disk.
## Bug Fixes and improvements
- Fix Scalar incorrect wall time display - Fix issues where different Python libraries conflicted in Mac
- Fix incorrect timestamp label display - Update Visual DL new logo
- Fix the Travis-CI not failing issue
- Fix the last recording not showing
- Update and clarify ImageWriter API docs for input params.
- VisualDL will call `SyncToDisk` before exiting.
- Adjust the interval between each `SyncToDisk` to improve performance.