Skip to content

ashar6194/velo_from_video

Repository files navigation

Building Model to estimate Velocity of Car from Front Cam Video

A pair of subsequent frames is selected to obtain optical flow. This flow image is used to learn the velocity of the ego-car.

Requirements-

python 3.5.2 Tensorflow-Gpu==1.7 Keras==2.2.4

File Info as follows-

  1. tdg.py -> Has Training Data Generator Object for Neural Networks Train/Eval
  2. data_creator.py -> Has functions to preprocess videos and visualize results on video
  3. utils.py -> Side utility functions useful in multiple parts of the repository
  4. io_args.py -> Argument Parser for different inputs
  5. train_eval.py -> Performs training
  6. eval_pipeline.py -> Store model's inferred output on train/test data into pickle files
  7. rough_plots.py -> Visualize model results and store results as text files
  8. model_set.py -> Keras Model Definitions

Results on Train+Val

alt text

A mini demo-

Releases

No releases published

Packages

No packages published

Languages