debug option

inference_test.py

@@ -13,40 +13,13 @@ import fnmatch
 import glob
 import subprocess
 from collections import Counter
-
+import matplotlib.pyplot as plt
 import re
 import numpy as np
 
 WATCH_DIR = "/tmp/mote-ocr-screenshots/"
 
 
-def deconvolve_image_path(input_path, psf_size=5, iterations=10, output_path=None):
-    """
-    Reads the image at input_path, applies RL deconvolution, and
-    returns an 8‑bit numpy array. Optionally saves to output_path.
-    """
-    img = io.imread(input_path)
-    if img.ndim == 3:
-        gray = color.rgb2gray(img)
-    else:
-        gray = img_as_float(img)
-    psf = np.ones((psf_size, psf_size), dtype=float)
-    psf /= psf.sum()
-    deconv = richardson_lucy(gray, psf, iterations=iterations)
-    deconv_u8 = img_as_ubyte(np.clip(deconv, 0, 1))
-    if output_path:
-        io.imsave(output_path, deconv_u8)
-    return deconv_u8
-
-def deconvolve_image_array(gray_float, psf_size=5, iterations=10):
-    """
-    Takes a grayscale float image in [0,1], runs RL deconv, and
-    returns an 8‑bit numpy array.
-    """
-    psf = np.ones((psf_size, psf_size), dtype=float)
-    psf /= psf.sum()
-    deconv = richardson_lucy(gray_float, psf, iterations=iterations)
-    return img_as_ubyte(np.clip(deconv, 0, 1))
 
 def contrast_stretch(gray):
     p2, p98 = np.percentile(gray, (2, 98))
@@ -173,13 +146,17 @@ def preprocess_image(image_path):
     gray = enhance_text(image_path)
     resampled = bicubic(gray) 
     # Convert grayscale to 3-channel RGB by duplicating the gray channel
-    img = contrast_stretch(adjust_gamma(cv2.cvtColor(resampled, cv2.COLOR_GRAY2RGB),gamma=1,alpha=1.65,beta=-1)) # perfect tunning
+    img = contrast_stretch(adjust_gamma(cv2.cvtColor(resampled, cv2.COLOR_GRAY2RGB),gamma=1.68,alpha=1.75,beta=-5)) # perfect tunning
+    plt.imshow(img)
+    plt.axis("off")
+    plt.savefig("/tmp/output.png", bbox_inches="tight", pad_inches=0)
+
     mat_in = ncnn.Mat.from_pixels_resize(
         img, ncnn.Mat.PixelType.PIXEL_RGB, img.shape[1], img.shape[0], 224, 224
     )
     
-    mean_vals = [0.5, 0.5, 0.5] * 4
-    norm_vals = [1/255, 1/255, 1/255] * 4
+    mean_vals = [0.5] * 3
+    norm_vals = [1/255] * 3
     mat_in.substract_mean_normalize(mean_vals, norm_vals)
     return mat_in