Download source and data from download page, open ./Face_Detection.ipynb and do task. It's easy.
Maximum mark for this task is 10 points:
- Prepare data (1 points)
- Student extracted positive and negative samples from data.
- Classifier training (3 points)
- Student add into model some layers.
- Student ran fitting and validation accuracy exceeded 90%.
- Student selected epoch with best validation loss and loaded this epoch weight.
- FCNN model (2 points)
- Student wrote fcnn model,
copy_weight
function and visualized activation heat map.
- Student wrote fcnn model,
- Detector (1 point)
- Student wrote
get_bboxes_and_decision_function
and visualized predicted bboxes
- Student wrote
- Precision/recall curve (1 point)
- Student implements precision/recall curve and plotted it.
- Threshold (1 point)
- Student find point for recall 0.85
- Precision/recall graph should stop at recall=0.85
- Detector score (1 point)
- On test dataset detection score (in graph header) should be 0.85 or greater.
This repository consist of multiple files:
Face_Detection.ipynb
-- main task, read and do.get_data.py
-- script to download data for task, run automatically from main task. You don't need download data manually.scores.py
-- scores, which are using in main task.graph.py
-- graph plotting and image showing functions.prepare_data.ipynb
-- prepare data to train and test, you may run this script and repeat learning-test procedure to make sure that model haven't over-fitting.
Dataset, used in this task is processed FDDB dataset. Processing explained in ./Face_Detection.ipynb and defined in ./prepare_data.ipynb
- Prepared by Vladimir Lutov: github.com/vslutov, vladimir.lutov@graphics.cs.msu.ru