donttap reader

#!/usr/bin/env python3

# https://youtu.be/vNIuTMhAp_Q

import ctypes
ctypes.windll.user32.SetProcessDPIAware()

import cv2
import numpy as np


# osu window
or_w, or_h = 512, 384
hist = np.zeros((or_h, or_w, 1))
coord_hist = {}

cap = cv2.VideoCapture('frenzy.mp4')
f = open('osu.txt', 'w')
buffer = ''
frames = cap.get(cv2.CAP_PROP_FRAME_COUNT)


# main loop
init = True
win = 'img'
cv2.namedWindow(win)
while True:
    cur_pos = cap.get(cv2.CAP_PROP_POS_FRAMES)
    cur_time = cap.get(cv2.CAP_PROP_POS_MSEC)
    print('%d/%d' % (cur_pos, frames), end='\r')
    ret, img = cap.read()
    if cur_pos < 228:
        continue
    if not ret or cv2.getWindowProperty(win, 0) < 0 or cur_pos > 2100:
        break

    img = cv2.resize(img, None, fx=0.8, fy=0.8)  # to make things easier
    img_h, img_w, _ = img.shape


    # ROI cutout
    x0, y0 = 321, 96
    x1, y1 = 321 + 382, 96 + 385
    roi_w, roi_h = x1 - x0, y1 - y0
    img = img[y0:y1, x0:x1]
    img_orig = img.copy()


    # image thresholding
    th_vals = [52, 43, 40]  # B, G, R
    img = cv2.GaussianBlur(img, (21, 21), 4)

    img_s = cv2.split(img)
    for i, ch in enumerate(img_s):
        ret, ch = cv2.threshold(ch, th_vals[i] - 2, 0, cv2.THRESH_TOZERO)
        ret, ch = cv2.threshold(ch, th_vals[i] + 2, 0, cv2.THRESH_TOZERO_INV)
        ret, ch = cv2.threshold(ch, th_vals[i] - 2, 255, cv2.THRESH_BINARY)
        img_s[i] = ch
    img = cv2.merge(img_s)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    grid = cv2.resize(gray, (4, 4), interpolation=cv2.INTER_AREA)
    ret, grid = cv2.threshold(grid, 128, 255, cv2.THRESH_BINARY)
    grid_copy = grid.copy()
    grid = grid.astype('int') / 255


    # Error check
    grid_active = grid[grid > 0]
    if grid_active.shape[0] != 3:
        print('ERROR ON FRAME %d' % cur_pos)
        exit()


    # calc osu coord and generate osu text
    if init:
        init_time = cur_time
        prev_grid = grid.copy()
        init = False
        continue
    else:
        diff_time = cur_time - init_time
        diff = grid - prev_grid
        row, col = np.where(diff < 0)

        if len(row) > 0:
            xo, yo = 98, 26
            wo, ho = 416 - xo, 345 - yo

            osu_xcoord = int(round(xo + col[0] * wo / 4 + wo / 8))
            osu_ycoord = int(round(yo + row[0] * ho / 4 + ho / 8))
            text = '%d,%d,%d,5,0,0:0:0:0:\n' % (osu_xcoord, osu_ycoord, diff_time)
            buffer += text

            osu_coord = (osu_xcoord, osu_ycoord)
            if osu_coord in coord_hist:
                coord_hist[osu_coord] += 1
            else:
                coord_hist[osu_coord] = 1

        prev_grid = grid.copy()


    # show current image
    gray = cv2.resize(grid_copy, (roi_w, roi_h), interpolation=cv2.INTER_NEAREST)
    cv2.imshow('grid', gray)
    cv2.imshow(win, img_orig)
    cv2.waitKey(1)


# clean up and save file
cv2.destroyAllWindows()
cap.release()
f.write(buffer)
f.close()


# view written osu coord hist.
def putTextCenter(img, text, pos, font, size, color, thickness, line=cv2.LINE_4):
    textsize = cv2.getTextSize(text, font, size, thickness)[0]
    textX = round(pos[0] - textsize[0] / 2)
    textY = round(pos[1] + textsize[1] / 2)
    cv2.putText(img, text, (textX, textY), font, size, color, thickness, line)

for key in coord_hist.keys():
    coord_count = coord_hist[key]
    putTextCenter(hist, '%d'%coord_count, key, cv2.FONT_HERSHEY_SIMPLEX, 0.4, 255, 1)
cv2.imshow('hist', hist)
cv2.waitKey(0)