Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
i-robot 9f1ed9e2cc | 1 year ago | |
---|---|---|
.. | ||
imgs | 1 year ago | |
src | 1 year ago | |
example.ipynb | 1 year ago | |
readme.md | 1 year ago |
The Lite-HRNet(Lite-HighResolutionNetwork) is a light weight backbone for high-resolution dependent computer vision tasks like Semantic Segmentation and Pose Estimation. It is a lightweight variant of HRNet(HighResolutionalNetwork), in which the authors replaces 1*1 convolutional blocks in HRNet with their innovative Cross Channel Weighting(CCW) to reduce both the time and spatial complexity.
In this case, we implement the Lite-HRNet using Mindspore for the task of human pose estimation.
On COCO2017:
Arch | Input Size | #Params | FLOPs | AP | AP50 | AP75 | AR | AR50 | ckpt |
---|---|---|---|---|---|---|---|---|---|
Naive Lite-HRNet-18 | 256x192 | 0.7M | 194.8M | 0.608 | 0.873 | 0.683 | 0.645 | 0.886 | Mindspore Download |
Wider Naive Lite-HRNet-18 | 256x192 | 1.3M | 311.1M | 0.643 | 0.893 | 0.725 | 0.680 | 0.901 | Mindspore Download |
Lite-HRNet-18 | 256x192 | 1.1M | 205.2M | 0.624 | 0.884 | 0.703 | 0.663 | 0.895 | Mindspore Download |
Lite-HRNet-18 | 384x288 | 1.1M | 461.6M | 0.668 | 0.895 | 0.739 | 0.702 | 0.909 | Mindspore Download |
Lite-HRNet-30 | 256x192 | 1.8M | 319.2M | 0.655 | 0.895 | 0.737 | 0.694 | 0.908 | Mindspore Download |
Lite-HRNet-30 | 384x288 | 1.8M | 717.8M | 0.698 | 0.905 | 0.781 | 0.731 | 0.918 | Mindspore Download |
On MPII:
Arch | Input Size | #Params | FLOPs | Mean | Mean@0.1 | ckpt |
---|---|---|---|---|---|---|
Naive Lite-HRNet-18 | 256x192 | 0.7M | 194.8M | 0.842 | 0.270 | Mindspore Download |
Wider Naive Lite-HRNet-18 | 256x256 | 1.3M | 311.1M | 0.859 | 0.288 | Mindspore Download |
Lite-HRNet-18 | 256x256 | 1.1M | 205.2M | 0.842 | 0.262 | Mindspore Download |
Lite-HRNet-30 | 256x256 | 1.8M | 319.2M | 0.861 | 0.280 | Mindspore Download |
You have to download COCO2017 dataset and annotation in COCODataset and make sure that the your path is organized as following:
Lite-HRNet/
├── imgs
├── src
├── annotations
├──person_keypoints_train2017.json
└──person_keypoints_train2017.json
├── train2017
└── val2017
python src/train.py
Optional Args | Default Value | Explanation |
---|---|---|
--model_type | lite_18 | The type of model |
--target_res | 256x192 | Resolution of resized input images |
--checkpoint_path | ./ckpts | Where to save or load checkpoints |
--train_batch | 32 | Training batch size |
--save_checkpoint_steps | 500 | The interval of saving checkpoints |
--load_ckpt | False | Load a checkpoint and continue training |
python src/eval.py
Optional Args | Default Value | Explanation |
---|---|---|
--model_type | lite_18 | The type of model |
--target_res | 256x192 | Resolution of resized input images |
--checkpoint_path | ./ckpts | Where to load checkpoints |
--output_path | ./eval_result | Where to save the evaluate result json file |
python src/eval.py
Optional Args | Default Value | Explanation |
---|---|---|
--model_type | lite_18 | The type of model |
--target_res | 256x192 | Resolution of resized input images |
--checkpoint_path | ./ckpts | Where to load checkpoints |
--infer_data_root | ./infer_data | Where to load the input data |
--out_data_root | ./out_data | Where to save the output data |
MindSpore实验,仅用于教学或培训目的。配合MindSpore官网使用。 MindSpore experiments, for teaching or training purposes only. Use it together with the MindSpore official website.
CSV Jupyter Notebook Text Python Markdown other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》