https://github.com/modeltc/lpcv_2023_solution
https://github.com/modeltc/lpcv_2023_solution
Last synced: 4 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/modeltc/lpcv_2023_solution
- Owner: ModelTC
- Created: 2023-08-10T09:36:29.000Z (over 2 years ago)
- Default Branch: master
- Last Pushed: 2023-11-29T06:40:44.000Z (about 2 years ago)
- Last Synced: 2025-09-20T23:51:27.571Z (5 months ago)
- Language: Python
- Size: 40.6 MB
- Stars: 18
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# LPCV 2023 solution
## LPCV 2023 introduction
https://lpcv.ai/2023LPCVC/introduction
## LPCV 2023 Leaderboard
https://lpcv.ai/scoreboard/Segmentation23
## News
- 10/31/2023 IEEE Computer Society published [a blog](https://www.computer.org/publications/tech-news/events/computer-vision-for-disaster-responses?source=home) about the LPCV 2023. Our team ModelTC was announced as the winner of this year’s competition.
## Inference Environment
### Hardware
[NVIDIA Jetson Nano 2GB](https://developer.nvidia.com/embedded/learn/get-started-jetson-nano-2gb-devkit)
### Software
refer to official sample: [23LPCVC_Segmentation_Track-Sample_Solution](https://github.com/lpcvai/23LPCVC_Segmentation_Track-Sample_Solution)
## Our submission
- file `submit_pyz/submit_0803.pyz`
- SHA256 = f1db90947eebedc3229bd5dd70ce5af586893ed8e370acce4f1b9ce33c62c315
- Submitted at **2023-08-03 09:16:28 EST**
- Perfomance Score **75.608**
## Usage of our inference code
- train a segmentation model on **your server**
- export a '.onnx' file from your framework on **your server**
- Use `trtexec` on **Jetson Nano**, convert onnx model to tensorrt model:
```shell
trtexec --workspace=4096 --onnx=xxx.onnx --saveEngine=xxx.trt --best --useSpinWait --outputIOFormats=fp16:chw --inputIOFormats=fp16:chw --verbose
```
- put the tensorrt model file into `inference_code` folder
- pack `inference_code` folder into pyz and submit(this step is same as the sample solution)
## Training
Goto [Train readme](./train/README.md) for training details, log, codebase, trained weights and reproduction