1. Improve model accuracy/performance to achieve real-time inference on edge
device.
2. Iterate model to cover all kinds of common features on highway/city scene.
3. Develop a single model to simultaneously output object detection and
segmentation result.
4. Apply model distillation, pruning, factorization, and quantization.
5. Data cleaning and augmentation.
6. Port to edge device, profile and find bottlenecks.
7. SOTA paper reading and implement, model zoo building.

1. Experience in CNN model training, especially on object detection and semantic segmentation.
2. Familiar with cost function rewritten and hyper parameters tuning.
3. Hands-on experience in deep learning framework, e.g., Pytorch, Tensorflow or Caffe.
4. Solid experience in system design and programming, ability to work with big data with minimal engineering support.