mkdir -p /tmp/workspace/histology-pdf && ls /tmp/workspace/histology-pdf/

cd /tmp/workspace/histology-pdf && curl -s -o source.jpg "https://d361swluvexzqs.cloudfront.net/chats-attachments/1000149870-1786267364369-510oj3.jpg" && python3 -c " from PIL import Image img = Image.open('source.jpg') print('Size:', img.size) print('Mode:', img.mode) "

cd /tmp/workspace/histology-pdf && python3 << 'EOF' from PIL import Image import numpy as np img = Image.open('source.jpg') w, h = img.size print(f"Image: {w} x {h}") # The image has 6 columns and rows: # Row 1: 1-6 (6 images) # Row 2: 7-12 (6 images) # Row 3: 13-18 (6 images) # Row 4: 19-24 (6 images) # Row 5: 25-31 (7 images) -- wait, 31 = 5*6 + 1 => last row has 1 but that can't be right # Actually 31 images: 5 rows of 6 = 30, plus 1 in row 6? No... # Let me look: rows 1-5 = 30 images (5x6), plus row 5 last = 31 # Actually from the image: rows 1-4 have 6 images each = 24, row 5 has 7? No... # From description: 6 cols, 5 rows but row 5 has only 6? That's 30. But there are 31. # Let me check: 6+6+6+6+7 = 31? Or 6+6+6+6+6+1? # Looking at the image more carefully: # Row 1: 1-6 # Row 2: 7-12 # Row 3: 13-18 # Row 4: 19-24 # Row 5: 25-30 (6 images) # Row 6: 31 (1 image)... unlikely # Actually it's more likely that the 5 rows have layouts of 6,6,6,6,7 or 6+6+6+6+7=31 # Let me just analyze pixel columns to find the grid arr = np.array(img) # Look at horizontal means to find column separators col_mean = arr.mean(axis=(0,2)) # mean over rows and channels -> per-column print("Image shape:", arr.shape) # Try to find white separators (high brightness columns) white_cols = np.where(col_mean > 240)[0] print("White col positions (first 20):", white_cols[:20] if len(white_cols) > 0 else "none") print("Total white cols:", len(white_cols)) EOF

cd /tmp/workspace/histology-pdf && python3 << 'EOF' from PIL import Image import numpy as np import random from reportlab.lib.pagesizes import A4 from reportlab.lib.units import mm from reportlab.platypus import SimpleDocTemplate, Image as RLImage, Spacer from reportlab.lib import colors import io, os img = Image.open('source.jpg') w, h = img.size arr = np.array(img) # The source image is 1536 x 1024 with: # 6 columns of images, each ~256px wide (1536/6=256) # 5 rows... but 31 images # Row 5 must have 7 images? No, 6*5=30, 6*4+7=31 # OR maybe it's: rows 1-4 have 6 cols = 24, row 5 has 7 but that doesn't divide evenly # 1536 / 7 = 219.4... not clean # Actually let me look: last row shows 25,26,27,28,29,30,31 = 7 images but image width / 7 = 219 # More likely 6 cols all the way, and row 5 has 6 images (25-30), row 6 has just 1 (31) # But checking image height: 1024 px, with label text at bottom of each cell # let me find row separators row_mean = arr.mean(axis=(1,2)) # per row, mean over cols and channels white_rows = np.where(row_mean > 230)[0] # Group contiguous white rows groups = [] if len(white_rows) > 0: start = white_rows[0] prev = white_rows[0] for r in white_rows[1:]: if r - prev > 1: groups.append((start, prev)) start = r prev = r groups.append((start, prev)) print("White row groups (separators):", groups) print("Row means at key positions:") for y in range(0, 1024, 50): print(f" y={y}: mean={row_mean[y]:.1f}") EOF

cd /tmp/workspace/histology-pdf && python3 << 'EOF' from PIL import Image import numpy as np import random, os from reportlab.lib.pagesizes import A4 from reportlab.platypus import SimpleDocTemplate, Image as RLImage from reportlab.lib.units import mm import io img = Image.open('source.jpg') w, h = img.size # 1536 x 1024 # From analysis: # Row separators (top of label text / white bands): # Row 1 images: y=13 to y=176 (image part only, skip label) # Labels appear ~y=177-209, row 2 starts ~y=210 etc. # Let's define image-only crop rows (exclude the label text below each photo): # Row 1: ~13 to ~176 # Row 2: ~210 to ~367 # Row 3: ~400 to ~557 # Row 4: ~590 to ~748 # Row 5: ~781 to ~961 # For columns: 6 images across 1536px # Col width ~ 256px each # Slight margin at edges row_tops = [13, 210, 400, 590, 781] row_bottoms = [176, 367, 557, 748, 961] # 6 columns col_lefts = [15, 271, 527, 783, 1039, 1295] col_widths = 248 # ~256 - small margin # Build grid: rows 1-5, 6 cols each = 30, plus row 5 only goes to col 1 for item 31 # Actually 31 items: rows 1-4 = 24, row 5 = 7 items # But image is 6 cols... let me check: 25-31 = 7 items in last row? That would be 7 cols # However the image is evenly divided into 6 cols (1536/6=256) # Let me check if last row really has 7 cols by examining actual tile widths # Names from image: rows 1-5 names = [ # Row 1 "Giant cell tumor", "Acute leukemia", "Fatty liver", "Cirrhosis", "Chronic myeloid leukemia", "Acute pyogenic meningitis", # Row 2 "Multiple myeloma", "Seminoma", "Atherosclerosis", "Thrombus", "Squamous cell carcinoma", "Macrocytic anemia", # Row 3 "Microcytic anemia", "Adenocarcinoma", "Peptic ulcer", "Caseous necrosis", "Myocardial infarction", "Acute appendicitis", # Row 4 "Keloid", "Acute pulmonary edema", "Septic infarct", "Adenoma", "Lipoma", "Basal cell carcinoma", # Row 5 (7 images) "Melanoma", "SLE (Lupus nephritis)", "Hashimoto thyroiditis", "Tuberculous granuloma", "Amyloidosis", "Hemosiderosis", "Rheumatoid nodule" ] # Row 5 has 7 images in 1536 width => each ~219.4 px # Let's define row5 cols separately row5_col_lefts = [] row5_col_width = 1536 // 7 # 219 for i in range(7): row5_col_lefts.append(i * row5_col_width) # Crop all tiles tiles = [] for row in range(4): for col in range(6): x1 = col_lefts[col] y1 = row_tops[row] x2 = x1 + col_widths y2 = row_bottoms[row] crop = img.crop((x1, y1, x2, y2)) tiles.append(crop) # Row 5: 7 images for col in range(7): x1 = row5_col_lefts[col] y1 = row_tops[4] x2 = x1 + row5_col_width y2 = row_bottoms[4] crop = img.crop((x1, y1, x2, y2)) tiles.append(crop) print(f"Total tiles: {len(tiles)}") # Save individual tiles for inspection os.makedirs('tiles', exist_ok=True) for i, t in enumerate(tiles): t.save(f'tiles/tile_{i+1:02d}.jpg') print("Tiles saved. Sizes:", tiles[0].size, tiles[24].size) EOF

cd /tmp/workspace/histology-pdf && python3 << 'EOF' from PIL import Image import numpy as np import random, os, io from reportlab.lib.pagesizes import A4 from reportlab.platypus import SimpleDocTemplate, Image as RLImage, Spacer from reportlab.lib.units import mm from reportlab.lib import colors from reportlab.pdfgen import canvas img_src = Image.open('source.jpg') w, h = img_src.size # 1536 x 1024 # ── Grid geometry ───────────────────────────────────────────────────────────── row_tops = [13, 210, 400, 590, 781] row_bottoms = [176, 367, 557, 748, 961] col_lefts = [15, 271, 527, 783, 1039, 1295] col_width = 248 row5_col_width = 1536 // 7 # 219 # ── Extract tiles ───────────────────────────────────────────────────────────── tiles = [] for row in range(4): for col in range(6): x1 = col_lefts[col] y1 = row_tops[row] x2 = x1 + col_width y2 = row_bottoms[row] tiles.append(img_src.crop((x1, y1, x2, y2))) for col in range(7): x1 = col * row5_col_width y1 = row_tops[4] x2 = x1 + row5_col_width y2 = row_bottoms[4] tiles.append(img_src.crop((x1, y1, x2, y2))) print(f"Extracted {len(tiles)} tiles") # ── Shuffle ──────────────────────────────────────────────────────────────────── random.seed(42) # reproducible shuffle indices = list(range(len(tiles))) random.shuffle(indices) print("Random order:", indices) # ── Build PDF one image per page ─────────────────────────────────────────────── page_w, page_h = A4 # 595 x 842 pt margin = 20 * mm # 20 mm margin on each side usable_w = page_w - 2*margin usable_h = page_h - 2*margin out_path = '/tmp/workspace/histology-pdf/histology_random.pdf' c = canvas.Canvas(out_path, pagesize=A4) for idx in indices: tile = tiles[idx] # Upscale to fill the page while keeping aspect ratio tw, th = tile.size scale = min(usable_w / tw, usable_h / th) draw_w = tw * scale draw_h = th * scale x = (page_w - draw_w) / 2 y = (page_h - draw_h) / 2 # Save tile to in-memory PNG for reportlab buf = io.BytesIO() tile_rgb = tile.convert('RGB') # Upscale tile for better quality new_tw = int(tw * scale) new_th = int(th * scale) tile_up = tile_rgb.resize((new_tw * 4, new_th * 4), Image.LANCZOS) tile_up.save(buf, format='PNG', optimize=True) buf.seek(0) from reportlab.lib.utils import ImageReader ir = ImageReader(buf) c.drawImage(ir, x, y, width=draw_w, height=draw_h) c.showPage() c.save() print(f"PDF saved: {out_path}") import os print(f"File size: {os.path.getsize(out_path)/1024:.1f} KB") EOF