geospiza-fortis/stab-swing-counter

counting frames

I run a template matching algorithm, and then pass the features into a multi-class classifier. Made a few templates based on a small region in the cropped screenshot for pattern matching in openCV. Here's 10 trials worth of data. The API changed since the previous one, where I'm using multiple smaller templates and omitting the position.

python main.py watch '.\data\v1\input\00.mkv' --crop --offset-x -175
python main.py prepare-samples '.\data\v1\input\00.mkv' data/v1/prep --offset-x -175 --frames 3600
python main.py train data/v1/labeled data/v1/model/logistic data/v1/templates
python main.py evaluate data/v1/prep data/v1/model_eval data/v1/model/logistic data/v1/templates

python main.py find-template '.\data\v1\input\00.mkv' '.\data\v1\templates\name.png'
# 932
python main.py find-template '.\data\v1\input\01.mkv' '.\data\v1\templates\name.png' --relative 932
# -69

python run_trials.py
python run_compress.py
python run_copy.py

# powershell uses backticks for continuations
python main.py train `
    data/v1/labeled `
    data/v1/prep `
    data/v1/model/logistic_sliding `
    data/v1/templates `
    --window 4 `
    --include-pos

python main.py evaluate `
    data/v1/prep `
    data/v1/model_eval_sliding `
    data/v1/model/logistic_sliding `
    data/v1/templates `
    --window 4 `
    --include-pos `
    --batch-size 60 `
    --input-type image

# powershell uses backticks for continuations
python main.py train `
    data/v1/labeled `
    data/v1/prep `
    data/v1/model/dt `
    data/v1/templates `
    --window 8 `
    --include-pos `
    --model decisiontree

python main.py evaluate `
    data/v1/prep `
    data/v1/model_eval_sliding `
    data/v1/model/dt `
    data/v1/templates `
    --window 8 `
    --include-pos `
    --batch-size 60 `
    --input-type image

# powershell uses backticks for continuations
python main.py train `
    data/v1/labeled `
    data/v1/prep `
    data/v1/model/dt `
    data/v1/templates `
    --window 8 `
    --no-include-pos `
    --model decisiontree

python main.py evaluate `
    data/v1/prep `
    data/v1/model_eval_sliding `
    data/v1/model/dt `
    data/v1/templates `
    --window 8 `
    --no-include-pos `
    --batch-size 60 `
    --input-type image

# powershell uses backticks for continuations
python main.py train `
    data/v1/labeled `
    data/v1/prep `
    data/v1/model/ridge `
    data/v1/templates `
    --window 4 `
    --no-include-pos `
    --model ridge

python main.py evaluate `
    data/v1/prep `
    data/v1/model_eval_sliding `
    data/v1/model/ridge `
    data/v1/templates `
    --window 4 `
    --no-include-pos `
    --batch-size 60 `
    --input-type image

This works great (I think).

python main.py evaluate `
    data/v1/input/01.mkv `
    data/v1/eval_01_ridge `
    data/v1/model/ridge `
    data/v1/templates `
    --window 4 `
    --no-include-pos `
    --batch-size 300 `
    --input-type video