|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- comma2k19 |
|
|
tags: |
|
|
- onnx |
|
|
- ENOT-AutoDL |
|
|
- ADAS |
|
|
--- |
|
|
|
|
|
# ENOT-AutoDL supercombo optimization on Comma2k19 dataset. |
|
|
|
|
|
This repository contains the modification for |
|
|
[supercombo](https://github.com/commaai/openpilot/tree/7d3ad941bc4ba4c923af7a1d7b48544bfc0d3e13/models) model |
|
|
with **one target task: plane prediction**. |
|
|
Model architecture from [openpilot-pipeline](https://github.com/mbalesni/openpilot-pipeline/blob/main/common/models/supercombo.onnx) repository is used as a baseline. |
|
|
Sample videos was generated using [openpilot-pipeline's code](https://github.com/mbalesni/openpilot-pipeline/blob/main/train/vis.ipynb). |
|
|
|
|
|
## Optimization results |
|
|
|
|
|
We use MACs as a latency measure because this metric is device-agnostic and implementation independent. |
|
|
There is also a possibility to optimize a model by target device latency using ENOT neural architecture selection algorithm. |
|
|
Please, keep in mind that acceleration by device latency differs from acceleration by MACs. |
|
|
|
|
|
| **Model** | **MACs** | **acceleration** | |
|
|
|---------------------------------|:---------:|:----------------:| |
|
|
| supercombo_plane.onnx | 472774912 | 1.0 | |
|
|
| optimized_supercombo_plane.onnx | 109917744 | 4.3 | |
|
|
|
|
|
## Inference example |
|
|
|
|
|
### Baseline model |
|
|
<img src="baseline.gif" width="640" height="480"/> |
|
|
|
|
|
### Optimized model |
|
|
<img src="accelerated.gif" width="640" height="480"/> |
|
|
|
|
|
If you want to book a demo, please contact us: [email protected] . |
