wantsjean / UISEE_AI_Competition_Baseline

UISEE Self-Driving Competition Baseline

UISEE_AI_Competition_Baseline

A simple baseline of UISEE Self-Driving Competition.

TODO List

• Add submit testing. training and submit by excute python submit.py
• Build a faster dataset, save images to npy files.
• Add more data augmentation, random resize, crop, rotate, ...
• Add pretraining datasets.

Getting Start

1. Git clone this repo.
2. Download the competition dataset from uisee.com & unzip it under data folder and rename it to uisee(for more datasets supporting)
3. Use tiff2npy to convert and resize tiff images to npy files
4. Baseline model is a simple 2D CNN + LSTM model, DIY your own model and adjust the Hyper Parameters in the config.py file, then start training.
5. this_repo/submit.txt will be generated by excuting submit.py.

Result on val set (split 20% from the traing set)

Epoch 10: Angle MAE = 0.94 Speed MAE = 1.76

Epoch 20: Angle MAE = 0.73 Speed MAE = 1.42