Use cv2.warpPerspective instead of dumb interpolation

https://docs.opencv.org/4.9.0/da/d6e/tutorial_py_geometric_transformations.html

3 files changed, 16 insertions, 43 deletions

diff --git a/README.md b/README.md
index 9857d2a..d465263 100644
--- a/README.md
+++ b/README.md
@@ -10,11 +10,11 @@ pip install --upgrade reedsolo --no-binary "reedsolo" --no-cache --config-settin
 
 ## Usage
 
-Encode: `python encoder.py -i in -o vid.mkv`
+Encode: `python encoder.py -i in`
 
 Play video: `mpv --scale=nearest --fullscreen --loop --no-keepaspect vid.mkv`
 
-Copy the flags printed by the encoder and pass them to the decoder: `python decoder.py -o out FLAGS`
+Copy the flags printed by the encoder and pass them to the decoder: `python decoder.py FLAGS`
 
 Formatting: `black -l 120 *.py`
 
diff --git a/decoder.py b/decoder.py
index d541d6e..2fb0183 100644
--- a/decoder.py
+++ b/decoder.py
@@ -1,15 +1,14 @@
 import argparse
-import sys
 import traceback
 import cv2
-import matplotlib.pyplot as plt
+# import matplotlib.pyplot as plt
 import numpy as np
 from creedsolo import RSCodec
 from raptorq import Decoder
 
 parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 parser.add_argument("-i", "--input", help="camera device index or input video file", default=0)
-parser.add_argument("-o", "--output", help="output file for decoded data")
+parser.add_argument("-o", "--output", help="output file for decoded data", default="out")
 parser.add_argument("-x", "--height", help="grid height", default=100, type=int)
 parser.add_argument("-y", "--width", help="grid width", default=100, type=int)
 parser.add_argument("-f", "--fps", help="frame rate", default=30, type=int)
@@ -38,7 +37,7 @@ def find_corner(A, f):
     mask = cv2.floodFill(
         A,
         np.empty(0),
-        tuple(reversed(guess)),
+        tuple(np.flip(guess)),
         0,
         (100, 100, 100),
         (100, 100, 100),
@@ -83,42 +82,16 @@ while data is None:
         bcol -= origin
         F = 255 * np.linalg.inv(np.stack((rcol, gcol, bcol)).T)
 
-        # Dumb perspective transform
-        xv = np.linspace(
-            -(cheight / 2 - 1) / (args.height - cheight + 1),
-            1 + (cheight / 2 - 1) / (args.height - cheight + 1),
-            args.height,
+        cch = cheight / 2 - 1
+        ccw = cwidth / 2 - 1
+        M = cv2.getPerspectiveTransform(
+            np.float32([np.flip(widx), np.flip(ridx), np.flip(gidx), np.flip(bidx)]),
+            np.float32([[ccw, cch], [args.width - ccw - 1, cch], [ccw, args.height - ccw - 1], [args.width - ccw - 1, args.height - cch - 1]]),
         )
-        yv = np.linspace(
-            -(cwidth / 2 - 1) / (args.width - cwidth + 1),
-            1 + (cwidth / 2 - 1) / (args.width - cwidth + 1),
-            args.width,
-        )
-        xp = (
-            np.outer(1 - xv, 1 - yv) * widx[0]
-            + np.outer(1 - xv, yv) * ridx[0]
-            + np.outer(xv, 1 - yv) * gidx[0]
-            + np.outer(xv, yv) * bidx[0]
-        )
-        yp = (
-            np.outer(1 - xv, 1 - yv) * widx[1]
-            + np.outer(1 - xv, yv) * ridx[1]
-            + np.outer(xv, 1 - yv) * gidx[1]
-            + np.outer(xv, yv) * bidx[1]
-        )
-
-        # plt.scatter(widx[1], widx[0])
-        # plt.scatter(ridx[1], ridx[0])
-        # plt.scatter(gidx[1], gidx[0])
-        # plt.scatter(bidx[1], bidx[0])
-        # plt.scatter(yp, xp)
-        # plt.imshow(raw_frame.astype(np.uint8))
-        # plt.show()
-
-        frame = raw_frame[
-            np.clip(np.round(xp).astype(np.int64), 0, X - 1), np.clip(np.round(yp).astype(np.int64), 0, Y - 1), :
-        ]
+        frame = cv2.warpPerspective(raw_frame, M, (args.width, args.height))
         frame = (np.squeeze(F @ (frame - origin)[..., np.newaxis]) >= 128).astype(np.uint8)
+        # plt.imshow(frame * 255)
+        # plt.show()
         frame = np.concatenate(
             (
                 frame[:cheight, cwidth : args.width - cwidth].flatten(),
@@ -130,7 +103,7 @@ while data is None:
         data = decoder.decode(bytes(rsc.decode(bytearray(np.packbits(frame) ^ frame_xor))[0][: args.psize]))
         print("Decoded frame")
     except KeyboardInterrupt:
-        sys.exit()
+        break
     except:
         traceback.print_exc()
 with open(args.output, "wb") as f:
diff --git a/encoder.py b/encoder.py
index c510b18..d80beae 100644
--- a/encoder.py
+++ b/encoder.py
@@ -6,7 +6,7 @@ from raptorq import Encoder
 
 parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 parser.add_argument("-i", "--input", help="input file")
-parser.add_argument("-o", "--output", help="output video file")
+parser.add_argument("-o", "--output", help="output video file", default="vid.mkv")
 parser.add_argument("-x", "--height", help="grid height", default=100, type=int)
 parser.add_argument("-y", "--width", help="grid width", default=100, type=int)
 parser.add_argument("-f", "--fps", help="frame rate", default=30, type=int)
@@ -39,7 +39,7 @@ rcorner = np.pad(np.dstack((ones, zeros, zeros)), ((0, 1), (1, 0), (0, 0)))
 gcorner = np.pad(np.dstack((zeros, ones, zeros)), ((1, 0), (0, 1), (0, 0)))
 bcorner = np.pad(np.dstack((zeros, zeros, ones)), ((1, 0), (1, 0), (0, 0)))
 
-print(f"-x {args.height} -y {args.width} -l {args.level} -s {len(data)} -p {len(packets[0])}")
+print(f"-x {args.height} -y {args.width} -l {args.level} -s {len(data)} -p {len(packets[0])}", end="")
 
 
 def frame(packet):