בלוג

בינה מלאכותית RB108-6T : תרגיל בית

תרגיל 1 : שימוש ב  GPU מול CPU

1.בדוק בעזרת בינה מלאכותית מה ההבדלים בין   GPU מול CPU

2.איזה GPU יש מי החברות השולטות ומה ההבדל TPU

3. איזה GPU יתאים לשואב רובוטי לבית , רחפן אוטונומי, למחשב לחישוב  קיפול חלבונים .

תרגיל 2 : תרגול עבודה עם GPU מול GPU

שים לב GPU שלך ב COLAB פעיל : T4

בחירה של GPU  בקולאב : runtime -> change runtime type

בחר  :

בחר :

תמונה לניסוי:

1. נתח את  הקוד הבא – נתח בעזרת
2. מה תפקיד הטנזור ?
3. מה תפקיד ספריית kornia
4. למה עדיף להשתמש ב kornia ולא OPENCV עבור טנזורים
5. הרץ את הקוד ב COLAB (לאחר הניתוח)
6. pip install kornia pillow torchvision matplotlib

   1	pip install kornia pillow torchvision matplotlib

   
    
7. # Google Colab - upload image, choose CPU/GPU (with fallback), BLUR -> CANNY, print timing in ms !pip -q install kornia pillow torchvision import os, time import torch import kornia as K from PIL import Image import torchvision.transforms as T import matplotlib.pyplot as plt from google.colab import files # ----------------------------- # PARAM: choose where to run # "auto" = use GPU if available # "cuda" = force GPU (fallback to CPU if not available) # "cpu" = force CPU # ----------------------------- RUN_ON = "cuda" # change to: "cpu" or "cuda" / "auto" has_cuda = torch.cuda.is_available() if RUN_ON == "cpu": device = "cpu" elif RUN_ON == "cuda": device = "cuda" if has_cuda else "cpu" else: # auto device = "cuda" if has_cuda else "cpu" print("Requested:", RUN_ON, "| CUDA available:", has_cuda, "| Using device:", device) def t_ms_sync(t0, device): if device == "cuda": torch.cuda.synchronize() return (time.perf_counter() - t0) * 1000.0 # ---- upload from your PC ---- uploaded = files.upload() img_path = next(iter(uploaded.keys())) print("Loaded:", img_path) # ---- timing: load + to tensor ---- t0 = time.perf_counter() img = Image.open(img_path).convert("RGB") x = T.ToTensor()(img).unsqueeze(0).to(device) # [1,3,H,W] load_ms = t_ms_sync(t0, device) # ---- timing: grayscale ---- t0 = time.perf_counter() gray = K.color.rgb_to_grayscale(x) # [1,1,H,W] gray_ms = t_ms_sync(t0, device) # ---- timing: blur ---- t0 = time.perf_counter() blur = K.filters.gaussian_blur2d(gray, kernel_size=(5, 5), sigma=(1.5, 1.5)) blur_ms = t_ms_sync(t0, device) # ---- timing: canny ---- t0 = time.perf_counter() edges, magnitude = K.filters.canny(blur, low_threshold=0.1, high_threshold=0.2) canny_ms = t_ms_sync(t0, device) # ---- timing: bring back + save ---- t0 = time.perf_counter() edges_u8 = (edges[0, 0].detach().clamp(0, 1).cpu() * 255).to(torch.uint8).numpy() out_path = os.path.splitext(img_path)[0] + f"_edges_{device}.png" Image.fromarray(edges_u8).save(out_path) save_ms = (time.perf_counter() - t0) * 1000.0 # CPU work print("\nTimings (ms):") print(f"Load + ToTensor + to(device): {load_ms:.3f} ms") print(f"Grayscale: {gray_ms:.3f} ms") print(f"Gaussian blur: {blur_ms:.3f} ms") print(f"Canny: {canny_ms:.3f} ms") print(f"CPU copy + save PNG: {save_ms:.3f} ms") print("Saved:", out_path) print("edges:", edges.shape, edges.device) # ---- plot ---- plt.figure(figsize=(6, 6)) plt.imshow(img); plt.axis("off"); plt.title("Original"); plt.show() plt.figure(figsize=(6, 6)) plt.imshow(edges_u8, cmap="gray"); plt.axis("off"); plt.title(f"Canny edges ({device})"); plt.show()

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83

   # Google Colab - upload image, choose CPU/GPU (with fallback), BLUR -> CANNY, print timing in ms !pip -q install kornia pillow torchvision   import os, time import torch import kornia as K from PIL import Image import torchvision.transforms as T import matplotlib.pyplot as plt from google.colab import files   # ----------------------------- # PARAM: choose where to run # "auto" = use GPU if available # "cuda" = force GPU (fallback to CPU if not available) # "cpu"  = force CPU # ----------------------------- RUN_ON = "cuda"   # change to: "cpu" or "cuda" / "auto"   has_cuda = torch.cuda.is_available()   if RUN_ON == "cpu":     device = "cpu" elif RUN_ON == "cuda":     device = "cuda" if has_cuda else "cpu" else:  # auto     device = "cuda" if has_cuda else "cpu"   print("Requested:", RUN_ON, "| CUDA available:", has_cuda, "| Using device:", device)   def t_ms_sync(t0, device):     if device == "cuda":         torch.cuda.synchronize()     return (time.perf_counter() - t0) * 1000.0   # ---- upload from your PC ---- uploaded = files.upload() img_path = next(iter(uploaded.keys())) print("Loaded:", img_path)   # ---- timing: load + to tensor ---- t0 = time.perf_counter() img = Image.open(img_path).convert("RGB") x = T.ToTensor()(img).unsqueeze(0).to(device)  # [1,3,H,W] load_ms = t_ms_sync(t0, device)   # ---- timing: grayscale ---- t0 = time.perf_counter() gray = K.color.rgb_to_grayscale(x)             # [1,1,H,W] gray_ms = t_ms_sync(t0, device)   # ---- timing: blur ---- t0 = time.perf_counter() blur = K.filters.gaussian_blur2d(gray, kernel_size=(5, 5), sigma=(1.5, 1.5)) blur_ms = t_ms_sync(t0, device)   # ---- timing: canny ---- t0 = time.perf_counter() edges, magnitude = K.filters.canny(blur, low_threshold=0.1, high_threshold=0.2) canny_ms = t_ms_sync(t0, device)   # ---- timing: bring back + save ---- t0 = time.perf_counter() edges_u8 = (edges[0, 0].detach().clamp(0, 1).cpu() * 255).to(torch.uint8).numpy() out_path = os.path.splitext(img_path)[0] + f"_edges_{device}.png" Image.fromarray(edges_u8).save(out_path) save_ms = (time.perf_counter() - t0) * 1000.0  # CPU work   print("\nTimings (ms):") print(f"Load + ToTensor + to(device): {load_ms:.3f} ms") print(f"Grayscale:                  {gray_ms:.3f} ms") print(f"Gaussian blur:              {blur_ms:.3f} ms") print(f"Canny:                      {canny_ms:.3f} ms") print(f"CPU copy + save PNG:        {save_ms:.3f} ms") print("Saved:", out_path) print("edges:", edges.shape, edges.device)   # ---- plot ---- plt.figure(figsize=(6, 6)) plt.imshow(img); plt.axis("off"); plt.title("Original"); plt.show()   plt.figure(figsize=(6, 6)) plt.imshow(edges_u8, cmap="gray"); plt.axis("off"); plt.title(f"Canny edges ({device})"); plt.show()

   
    
8. מה ההבדלים בין GPU ל CPU – בזמנים
9. שינוי קוד  : הוסף פילטר EMBUSS והצג גם אותו כתמונה