File size: 8,642 Bytes
eb09c29 |
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 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 |
#!/usr/bin/env python3
"""
Debug script to test and verify the tensor creation fix.
This script isolates the problematic code and tests various scenarios.
"""
import sys
import tempfile
from pathlib import Path
import logging
import numpy as np
from PIL import Image
# Configure detailed logging
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')
def create_test_frames(num_frames=8, size=(224, 224)):
"""Create synthetic test frames to simulate video processing."""
frames = []
for i in range(num_frames):
# Create a simple gradient image
img_array = np.zeros((*size, 3), dtype=np.uint8)
# Add some variation between frames
gradient = np.linspace(0, 255, size[0]).astype(np.uint8)
for j in range(3): # RGB channels
img_array[:, :, j] = gradient + (i * 10) % 256
# Convert to PIL Image
frame = Image.fromarray(img_array, 'RGB')
frames.append(frame)
return frames
def test_processor_approaches():
"""Test different approaches to fix the tensor creation issue."""
print("🔍 Testing Tensor Creation Fix")
print("=" * 50)
try:
from transformers import AutoImageProcessor, TimesformerForVideoClassification
import torch
except ImportError as e:
print(f"❌ Missing dependencies: {e}")
return False
# Load processor (but not full model to save time/memory)
try:
processor = AutoImageProcessor.from_pretrained("facebook/timesformer-base-finetuned-k400")
print("✅ Processor loaded successfully")
except Exception as e:
print(f"❌ Failed to load processor: {e}")
return False
# Test with different frame scenarios
test_scenarios = [
{"name": "Standard 8 frames", "frames": 8, "size": (224, 224)},
{"name": "Different count (6 frames)", "frames": 6, "size": (224, 224)},
{"name": "Different size frames", "frames": 8, "size": (256, 256)},
{"name": "Single frame", "frames": 1, "size": (224, 224)},
]
success_count = 0
for scenario in test_scenarios:
print(f"\n📋 Testing: {scenario['name']}")
print("-" * 30)
frames = create_test_frames(scenario["frames"], scenario["size"])
required_frames = 8 # TimeSformer default
# Apply the same logic as in our fix
if len(frames) != required_frames:
print(f"⚠️ Frame count mismatch: {len(frames)} vs {required_frames}")
if len(frames) < required_frames:
frames.extend([frames[-1]] * (required_frames - len(frames)))
print(f"🔧 Padded to {len(frames)} frames")
else:
frames = frames[:required_frames]
print(f"🔧 Truncated to {len(frames)} frames")
# Ensure consistent frame sizes
if frames:
target_size = (224, 224) # Standard size for TimeSformer
frames = [frame.resize(target_size) if frame.size != target_size else frame for frame in frames]
print(f"🔧 Normalized all frames to {target_size}")
# Test different processor approaches
approaches = [
("Direct with padding", lambda: processor(images=frames, return_tensors="pt", padding=True)),
("List wrapped with padding", lambda: processor(images=[frames], return_tensors="pt", padding=True)),
("Direct without padding", lambda: processor(images=frames, return_tensors="pt")),
("Manual tensor creation", lambda: create_manual_tensor(frames, processor)),
]
for approach_name, approach_func in approaches:
try:
print(f" 🧪 Trying: {approach_name}")
inputs = approach_func()
# Check tensor properties
if 'pixel_values' in inputs:
tensor = inputs['pixel_values']
print(f" ✅ Success! Tensor shape: {tensor.shape}")
print(f" 📊 Tensor dtype: {tensor.dtype}")
print(f" 📈 Tensor range: [{tensor.min():.3f}, {tensor.max():.3f}]")
success_count += 1
break
else:
print(f" ❌ No pixel_values in output: {inputs.keys()}")
except Exception as e:
print(f" ❌ Failed: {str(e)[:100]}...")
continue
else:
print(f" 💥 All approaches failed for {scenario['name']}")
print(f"\n📊 Summary: {success_count}/{len(test_scenarios)} scenarios passed")
return success_count == len(test_scenarios)
def create_manual_tensor(frames, processor):
"""Manual tensor creation as final fallback."""
if not frames:
raise ValueError("No frames provided")
frame_arrays = []
for frame in frames:
# Ensure RGB mode
if frame.mode != 'RGB':
frame = frame.convert('RGB')
# Resize to standard size
frame = frame.resize((224, 224))
frame_array = np.array(frame)
frame_arrays.append(frame_array)
# Stack frames: (num_frames, height, width, channels)
video_array = np.stack(frame_arrays)
# Convert to tensor and normalize
video_tensor = torch.tensor(video_array, dtype=torch.float32) / 255.0
# Rearrange dimensions for TimeSformer: (batch, channels, num_frames, height, width)
video_tensor = video_tensor.permute(3, 0, 1, 2).unsqueeze(0)
return {'pixel_values': video_tensor}
def test_video_processing():
"""Test with actual video processing simulation."""
print(f"\n🎬 Testing Video Processing Pipeline")
print("=" * 50)
try:
# Create a temporary "video" by saving frames as images
with tempfile.TemporaryDirectory() as tmp_dir:
tmp_path = Path(tmp_dir)
# Create test frames and save them
frames = create_test_frames(8, (640, 480)) # Different size to test resizing
frame_paths = []
for i, frame in enumerate(frames):
frame_path = tmp_path / f"frame_{i:03d}.jpg"
frame.save(frame_path)
frame_paths.append(frame_path)
print(f"✅ Created {len(frame_paths)} test frames")
# Load frames back (simulating video reading)
loaded_frames = []
for frame_path in frame_paths:
frame = Image.open(frame_path)
loaded_frames.append(frame)
print(f"✅ Loaded {len(loaded_frames)} frames")
# Test processing
return test_single_scenario(loaded_frames, "Video simulation")
except Exception as e:
print(f"❌ Video processing test failed: {e}")
return False
def test_single_scenario(frames, scenario_name):
"""Test a single scenario with comprehensive error handling."""
print(f"\n🎯 Testing scenario: {scenario_name}")
try:
from transformers import AutoImageProcessor
import torch
processor = AutoImageProcessor.from_pretrained("facebook/timesformer-base-finetuned-k400")
# Apply our fix logic
required_frames = 8
if len(frames) != required_frames:
if len(frames) < required_frames:
frames.extend([frames[-1]] * (required_frames - len(frames)))
else:
frames = frames[:required_frames]
# Normalize frame sizes
target_size = (224, 224)
frames = [frame.resize(target_size) if frame.size != target_size else frame for frame in frames]
# Try our primary approach
inputs = processor(images=frames, return_tensors="pt", padding=True)
print(f"✅ Success! Tensor shape: {inputs['pixel_values'].shape}")
return True
except Exception as e:
print(f"❌ Failed: {e}")
return False
if __name__ == "__main__":
print("🐛 Tensor Creation Debug Suite")
print("=" * 60)
# Test 1: Processor approaches
test1_passed = test_processor_approaches()
# Test 2: Video processing simulation
test2_passed = test_video_processing()
print(f"\n🏁 Final Results:")
print(f" Processor tests: {'✅ PASSED' if test1_passed else '❌ FAILED'}")
print(f" Video tests: {'✅ PASSED' if test2_passed else '❌ FAILED'}")
if test1_passed and test2_passed:
print(f"\n🎉 All tests passed! The tensor fix should work correctly.")
sys.exit(0)
else:
print(f"\n💥 Some tests failed. Check the logs above for details.")
sys.exit(1)
|