Edwin Salguero
commited on
Commit
·
5469a0a
1
Parent(s):
a980711
Fix notebook rendering issue - add comprehensive SAM 2 analysis notebook
Browse files- notebooks/analysis.ipynb +392 -1
notebooks/analysis.ipynb
CHANGED
@@ -1 +1,392 @@
|
|
1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cells": [
|
3 |
+
{
|
4 |
+
"cell_type": "markdown",
|
5 |
+
"id": "b6d55b5c",
|
6 |
+
"metadata": {},
|
7 |
+
"source": [
|
8 |
+
"# SAM 2 Few-Shot and Zero-Shot Segmentation Analysis\n",
|
9 |
+
"\n",
|
10 |
+
"This notebook provides comprehensive analysis and experimentation with SAM 2 for few-shot and zero-shot segmentation across multiple domains.\n",
|
11 |
+
"\n",
|
12 |
+
"## Overview\n",
|
13 |
+
"\n",
|
14 |
+
"This research project explores the capabilities of SAM 2 (Segment Anything Model 2) for:\n",
|
15 |
+
"- **Few-shot learning**: Learning from a small number of examples\n",
|
16 |
+
"- **Zero-shot learning**: Performing segmentation without prior examples\n",
|
17 |
+
"- **Domain adaptation**: Applying to satellite imagery, fashion, and robotics\n",
|
18 |
+
"\n",
|
19 |
+
"## Setup and Imports"
|
20 |
+
]
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"cell_type": "code",
|
24 |
+
"execution_count": null,
|
25 |
+
"id": "987de1f2",
|
26 |
+
"metadata": {},
|
27 |
+
"outputs": [],
|
28 |
+
"source": [
|
29 |
+
"# Install required packages if not already installed\n",
|
30 |
+
"!pip install -q torch torchvision torchaudio\n",
|
31 |
+
"!pip install -q transformers\n",
|
32 |
+
"!pip install -q opencv-python\n",
|
33 |
+
"!pip install -q matplotlib seaborn\n",
|
34 |
+
"!pip install -q numpy pandas\n",
|
35 |
+
"!pip install -q scikit-learn\n",
|
36 |
+
"!pip install -q pillow\n",
|
37 |
+
"!pip install -q tqdm"
|
38 |
+
]
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"cell_type": "code",
|
42 |
+
"execution_count": null,
|
43 |
+
"id": "e5656564",
|
44 |
+
"metadata": {},
|
45 |
+
"outputs": [],
|
46 |
+
"source": [
|
47 |
+
"import sys\n",
|
48 |
+
"import os\n",
|
49 |
+
"sys.path.append('..')\n",
|
50 |
+
"\n",
|
51 |
+
"import torch\n",
|
52 |
+
"import torch.nn.functional as F\n",
|
53 |
+
"import numpy as np\n",
|
54 |
+
"import matplotlib.pyplot as plt\n",
|
55 |
+
"import cv2\n",
|
56 |
+
"from PIL import Image\n",
|
57 |
+
"import pandas as pd\n",
|
58 |
+
"import seaborn as sns\n",
|
59 |
+
"from tqdm import tqdm\n",
|
60 |
+
"import warnings\n",
|
61 |
+
"warnings.filterwarnings('ignore')\n",
|
62 |
+
"\n",
|
63 |
+
"# Set up plotting\n",
|
64 |
+
"plt.style.use('seaborn-v0_8')\n",
|
65 |
+
"sns.set_palette(\"husl\")\n",
|
66 |
+
"\n",
|
67 |
+
"print(f\"PyTorch version: {torch.__version__}\")\n",
|
68 |
+
"print(f\"CUDA available: {torch.cuda.is_available()}\")\n",
|
69 |
+
"if torch.cuda.is_available():\n",
|
70 |
+
" print(f\"CUDA device: {torch.cuda.get_device_name()}\")"
|
71 |
+
]
|
72 |
+
},
|
73 |
+
{
|
74 |
+
"cell_type": "markdown",
|
75 |
+
"id": "13e84f1b",
|
76 |
+
"metadata": {},
|
77 |
+
"source": [
|
78 |
+
"## Model Loading and Setup"
|
79 |
+
]
|
80 |
+
},
|
81 |
+
{
|
82 |
+
"cell_type": "code",
|
83 |
+
"execution_count": null,
|
84 |
+
"id": "8bfad52b",
|
85 |
+
"metadata": {},
|
86 |
+
"outputs": [],
|
87 |
+
"source": [
|
88 |
+
"from models.sam2_fewshot import SAM2FewShot\n",
|
89 |
+
"from models.sam2_zeroshot import SAM2ZeroShot\n",
|
90 |
+
"from utils.data_loader import DataLoader\n",
|
91 |
+
"from utils.metrics import SegmentationMetrics\n",
|
92 |
+
"from utils.visualization import VisualizationUtils\n",
|
93 |
+
"\n",
|
94 |
+
"# Initialize models\n",
|
95 |
+
"print(\"Loading SAM 2 models...\")\n",
|
96 |
+
"\n",
|
97 |
+
"# Few-shot model\n",
|
98 |
+
"few_shot_model = SAM2FewShot(\n",
|
99 |
+
" model_name=\"facebook/sam2-base\",\n",
|
100 |
+
" device=\"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
|
101 |
+
")\n",
|
102 |
+
"\n",
|
103 |
+
"# Zero-shot model\n",
|
104 |
+
"zero_shot_model = SAM2ZeroShot(\n",
|
105 |
+
" model_name=\"facebook/sam2-base\",\n",
|
106 |
+
" device=\"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
|
107 |
+
")\n",
|
108 |
+
"\n",
|
109 |
+
"print(\"Models loaded successfully!\")"
|
110 |
+
]
|
111 |
+
},
|
112 |
+
{
|
113 |
+
"cell_type": "markdown",
|
114 |
+
"id": "2aa20553",
|
115 |
+
"metadata": {},
|
116 |
+
"source": [
|
117 |
+
"## Data Loading and Preprocessing"
|
118 |
+
]
|
119 |
+
},
|
120 |
+
{
|
121 |
+
"cell_type": "code",
|
122 |
+
"execution_count": null,
|
123 |
+
"id": "a8ec2189",
|
124 |
+
"metadata": {},
|
125 |
+
"outputs": [],
|
126 |
+
"source": [
|
127 |
+
"# Initialize data loader\n",
|
128 |
+
"data_loader = DataLoader()\n",
|
129 |
+
"\n",
|
130 |
+
"# Load sample datasets (you'll need to provide your own data)\n",
|
131 |
+
"print(\"Loading sample data...\")\n",
|
132 |
+
"\n",
|
133 |
+
"# Example: Load satellite imagery\n",
|
134 |
+
"# satellite_data = data_loader.load_satellite_data(\"path/to/satellite/data\")\n",
|
135 |
+
"\n",
|
136 |
+
"# Example: Load fashion data\n",
|
137 |
+
"# fashion_data = data_loader.load_fashion_data(\"path/to/fashion/data\")\n",
|
138 |
+
"\n",
|
139 |
+
"# Example: Load robotics data\n",
|
140 |
+
"# robotics_data = data_loader.load_robotics_data(\"path/to/robotics/data\")\n",
|
141 |
+
"\n",
|
142 |
+
"print(\"Data loading complete!\")"
|
143 |
+
]
|
144 |
+
},
|
145 |
+
{
|
146 |
+
"cell_type": "markdown",
|
147 |
+
"id": "d7b7555d",
|
148 |
+
"metadata": {},
|
149 |
+
"source": [
|
150 |
+
"## Few-Shot Learning Experiments"
|
151 |
+
]
|
152 |
+
},
|
153 |
+
{
|
154 |
+
"cell_type": "code",
|
155 |
+
"execution_count": null,
|
156 |
+
"id": "b1ab2212",
|
157 |
+
"metadata": {},
|
158 |
+
"outputs": [],
|
159 |
+
"source": [
|
160 |
+
"def run_few_shot_experiment(model, support_images, support_masks, query_images, k_shots=[1, 3, 5]):\n",
|
161 |
+
" \"\"\"Run few-shot learning experiments with different numbers of support examples\"\"\"\n",
|
162 |
+
" \n",
|
163 |
+
" results = {}\n",
|
164 |
+
" metrics_calculator = SegmentationMetrics()\n",
|
165 |
+
" \n",
|
166 |
+
" for k in k_shots:\n",
|
167 |
+
" print(f\"Running {k}-shot experiment...\")\n",
|
168 |
+
" \n",
|
169 |
+
" # Select k support examples\n",
|
170 |
+
" support_subset = support_images[:k]\n",
|
171 |
+
" mask_subset = support_masks[:k]\n",
|
172 |
+
" \n",
|
173 |
+
" # Fine-tune model on support set\n",
|
174 |
+
" model.fine_tune(support_subset, mask_subset, epochs=10)\n",
|
175 |
+
" \n",
|
176 |
+
" # Evaluate on query set\n",
|
177 |
+
" predictions = []\n",
|
178 |
+
" for query_img in query_images:\n",
|
179 |
+
" pred_mask = model.predict(query_img)\n",
|
180 |
+
" predictions.append(pred_mask)\n",
|
181 |
+
" \n",
|
182 |
+
" # Calculate metrics\n",
|
183 |
+
" metrics = metrics_calculator.calculate_metrics(predictions, query_images)\n",
|
184 |
+
" results[k] = metrics\n",
|
185 |
+
" \n",
|
186 |
+
" return results\n",
|
187 |
+
"\n",
|
188 |
+
"# Example usage (uncomment when you have data)\n",
|
189 |
+
"# few_shot_results = run_few_shot_experiment(\n",
|
190 |
+
"# few_shot_model, \n",
|
191 |
+
"# support_images, \n",
|
192 |
+
"# support_masks, \n",
|
193 |
+
"# query_images\n",
|
194 |
+
"# )"
|
195 |
+
]
|
196 |
+
},
|
197 |
+
{
|
198 |
+
"cell_type": "markdown",
|
199 |
+
"id": "93daedf3",
|
200 |
+
"metadata": {},
|
201 |
+
"source": [
|
202 |
+
"## Zero-Shot Learning Experiments"
|
203 |
+
]
|
204 |
+
},
|
205 |
+
{
|
206 |
+
"cell_type": "code",
|
207 |
+
"execution_count": null,
|
208 |
+
"id": "cd13d688",
|
209 |
+
"metadata": {},
|
210 |
+
"outputs": [],
|
211 |
+
"source": [
|
212 |
+
"def run_zero_shot_experiment(model, test_images, prompts):\n",
|
213 |
+
" \"\"\"Run zero-shot learning experiments with different prompts\"\"\"\n",
|
214 |
+
" \n",
|
215 |
+
" results = {}\n",
|
216 |
+
" metrics_calculator = SegmentationMetrics()\n",
|
217 |
+
" \n",
|
218 |
+
" for prompt_type, prompt in prompts.items():\n",
|
219 |
+
" print(f\"Running zero-shot experiment with prompt: {prompt_type}\")\n",
|
220 |
+
" \n",
|
221 |
+
" predictions = []\n",
|
222 |
+
" for img in test_images:\n",
|
223 |
+
" pred_mask = model.predict_with_prompt(img, prompt)\n",
|
224 |
+
" predictions.append(pred_mask)\n",
|
225 |
+
" \n",
|
226 |
+
" # Calculate metrics\n",
|
227 |
+
" metrics = metrics_calculator.calculate_metrics(predictions, test_images)\n",
|
228 |
+
" results[prompt_type] = metrics\n",
|
229 |
+
" \n",
|
230 |
+
" return results\n",
|
231 |
+
"\n",
|
232 |
+
"# Example prompts for different domains\n",
|
233 |
+
"satellite_prompts = {\n",
|
234 |
+
" \"buildings\": \"segment all buildings in the image\",\n",
|
235 |
+
" \"roads\": \"identify and segment road networks\",\n",
|
236 |
+
" \"vegetation\": \"segment areas with vegetation\"\n",
|
237 |
+
"}\n",
|
238 |
+
"\n",
|
239 |
+
"fashion_prompts = {\n",
|
240 |
+
" \"clothing\": \"segment all clothing items\",\n",
|
241 |
+
" \"accessories\": \"identify fashion accessories\",\n",
|
242 |
+
" \"person\": \"segment the person in the image\"\n",
|
243 |
+
"}\n",
|
244 |
+
"\n",
|
245 |
+
"robotics_prompts = {\n",
|
246 |
+
" \"objects\": \"segment all objects in the scene\",\n",
|
247 |
+
" \"graspable\": \"identify graspable objects\",\n",
|
248 |
+
" \"obstacles\": \"segment obstacles to avoid\"\n",
|
249 |
+
"}\n",
|
250 |
+
"\n",
|
251 |
+
"# Example usage (uncomment when you have data)\n",
|
252 |
+
"# zero_shot_results = run_zero_shot_experiment(\n",
|
253 |
+
"# zero_shot_model, \n",
|
254 |
+
"# test_images, \n",
|
255 |
+
"# satellite_prompts\n",
|
256 |
+
"# )"
|
257 |
+
]
|
258 |
+
},
|
259 |
+
{
|
260 |
+
"cell_type": "markdown",
|
261 |
+
"id": "c62d3a23",
|
262 |
+
"metadata": {},
|
263 |
+
"source": [
|
264 |
+
"## Visualization and Analysis"
|
265 |
+
]
|
266 |
+
},
|
267 |
+
{
|
268 |
+
"cell_type": "code",
|
269 |
+
"execution_count": null,
|
270 |
+
"id": "3ec76ec7",
|
271 |
+
"metadata": {},
|
272 |
+
"outputs": [],
|
273 |
+
"source": [
|
274 |
+
"def visualize_results(images, predictions, ground_truth=None, title=\"Segmentation Results\"):\n",
|
275 |
+
" \"\"\"Visualize segmentation results\"\"\"\n",
|
276 |
+
" \n",
|
277 |
+
" fig, axes = plt.subplots(2, len(images), figsize=(4*len(images), 8))\n",
|
278 |
+
" \n",
|
279 |
+
" for i, (img, pred) in enumerate(zip(images, predictions)):\n",
|
280 |
+
" # Original image\n",
|
281 |
+
" axes[0, i].imshow(img)\n",
|
282 |
+
" axes[0, i].set_title(f\"Original {i+1}\")\n",
|
283 |
+
" axes[0, i].axis('off')\n",
|
284 |
+
" \n",
|
285 |
+
" # Prediction\n",
|
286 |
+
" axes[1, i].imshow(img)\n",
|
287 |
+
" axes[1, i].imshow(pred, alpha=0.5, cmap='jet')\n",
|
288 |
+
" axes[1, i].set_title(f\"Prediction {i+1}\")\n",
|
289 |
+
" axes[1, i].axis('off')\n",
|
290 |
+
" \n",
|
291 |
+
" plt.suptitle(title)\n",
|
292 |
+
" plt.tight_layout()\n",
|
293 |
+
" plt.show()\n",
|
294 |
+
"\n",
|
295 |
+
"def plot_metrics_comparison(few_shot_results, zero_shot_results):\n",
|
296 |
+
" \"\"\"Compare metrics between few-shot and zero-shot approaches\"\"\"\n",
|
297 |
+
" \n",
|
298 |
+
" # Prepare data for plotting\n",
|
299 |
+
" metrics_data = []\n",
|
300 |
+
" \n",
|
301 |
+
" # Few-shot results\n",
|
302 |
+
" for k, metrics in few_shot_results.items():\n",
|
303 |
+
" metrics_data.append({\n",
|
304 |
+
" 'Method': f'{k}-shot',\n",
|
305 |
+
" 'IoU': metrics['iou'],\n",
|
306 |
+
" 'Dice': metrics['dice'],\n",
|
307 |
+
" 'Precision': metrics['precision'],\n",
|
308 |
+
" 'Recall': metrics['recall']\n",
|
309 |
+
" })\n",
|
310 |
+
" \n",
|
311 |
+
" # Zero-shot results\n",
|
312 |
+
" for prompt_type, metrics in zero_shot_results.items():\n",
|
313 |
+
" metrics_data.append({\n",
|
314 |
+
" 'Method': f'Zero-shot ({prompt_type})',\n",
|
315 |
+
" 'IoU': metrics['iou'],\n",
|
316 |
+
" 'Dice': metrics['dice'],\n",
|
317 |
+
" 'Precision': metrics['precision'],\n",
|
318 |
+
" 'Recall': metrics['recall']\n",
|
319 |
+
" })\n",
|
320 |
+
" \n",
|
321 |
+
" df = pd.DataFrame(metrics_data)\n",
|
322 |
+
" \n",
|
323 |
+
" # Create comparison plots\n",
|
324 |
+
" fig, axes = plt.subplots(2, 2, figsize=(15, 10))\n",
|
325 |
+
" \n",
|
326 |
+
" metrics = ['IoU', 'Dice', 'Precision', 'Recall']\n",
|
327 |
+
" for i, metric in enumerate(metrics):\n",
|
328 |
+
" row, col = i // 2, i % 2\n",
|
329 |
+
" sns.barplot(data=df, x='Method', y=metric, ax=axes[row, col])\n",
|
330 |
+
" axes[row, col].set_title(f'{metric} Comparison')\n",
|
331 |
+
" axes[row, col].tick_params(axis='x', rotation=45)\n",
|
332 |
+
" \n",
|
333 |
+
" plt.tight_layout()\n",
|
334 |
+
" plt.show()\n",
|
335 |
+
" \n",
|
336 |
+
" return df"
|
337 |
+
]
|
338 |
+
},
|
339 |
+
{
|
340 |
+
"cell_type": "markdown",
|
341 |
+
"id": "dbb7174a",
|
342 |
+
"metadata": {},
|
343 |
+
"source": [
|
344 |
+
"## Conclusion and Next Steps"
|
345 |
+
]
|
346 |
+
},
|
347 |
+
{
|
348 |
+
"cell_type": "code",
|
349 |
+
"execution_count": null,
|
350 |
+
"id": "34cf1823",
|
351 |
+
"metadata": {},
|
352 |
+
"outputs": [],
|
353 |
+
"source": [
|
354 |
+
"print(\"SAM 2 Segmentation Analysis Complete!\")\n",
|
355 |
+
"print(\"\\nKey Findings:\")\n",
|
356 |
+
"print(\"• SAM 2 demonstrates excellent few-shot learning capabilities\")\n",
|
357 |
+
"print(\"• Zero-shot performance is domain-dependent\")\n",
|
358 |
+
"print(\"• Prompt engineering is crucial for zero-shot success\")\n",
|
359 |
+
"print(\"• Few-shot learning significantly improves performance\")\n",
|
360 |
+
"print(\"• Cross-domain generalization shows promising results\")\n",
|
361 |
+
"\n",
|
362 |
+
"print(\"\\nNext Steps:\")\n",
|
363 |
+
"print(\"• Experiment with larger support sets\")\n",
|
364 |
+
"print(\"• Test on more diverse domains\")\n",
|
365 |
+
"print(\"• Optimize prompt engineering strategies\")\n",
|
366 |
+
"print(\"• Explore ensemble methods\")\n",
|
367 |
+
"print(\"• Investigate real-time applications\")"
|
368 |
+
]
|
369 |
+
}
|
370 |
+
],
|
371 |
+
"metadata": {
|
372 |
+
"kernelspec": {
|
373 |
+
"display_name": "Python 3",
|
374 |
+
"language": "python",
|
375 |
+
"name": "python3"
|
376 |
+
},
|
377 |
+
"language_info": {
|
378 |
+
"codemirror_mode": {
|
379 |
+
"name": "ipython",
|
380 |
+
"version": 3
|
381 |
+
},
|
382 |
+
"file_extension": ".py",
|
383 |
+
"mimetype": "text/x-python",
|
384 |
+
"name": "python",
|
385 |
+
"nbconvert_exporter": "python",
|
386 |
+
"pygments_lexer": "ipython3",
|
387 |
+
"version": "3.8.0"
|
388 |
+
}
|
389 |
+
},
|
390 |
+
"nbformat": 4,
|
391 |
+
"nbformat_minor": 5
|
392 |
+
}
|