File size: 13,026 Bytes
d3c0eea | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 | #!/usr/bin/env python3
"""
CFP-Jarvis1 Streaming Mode - Real-time screen analysis and action
"""
import os
import time
import threading
import queue
import torch
import numpy as np
from PIL import Image
import mss
import cv2
from transformers import AutoProcessor, AutoModelForImageTextToText
from dataclasses import dataclass
from typing import Optional, Tuple
import pyautogui
@dataclass
class Action:
"""Represents an action to perform"""
type: str # 'click', 'type', 'key', 'move'
coordinates: Optional[Tuple[int, int]] = None
text: Optional[str] = None
confidence: float = 0.0
class CFPJarvis1Stream:
"""
CFP-Jarvis1 with streaming capabilities
Analyzes screen in real-time and performs actions
"""
def __init__(self,
model_path="OpenGVLab/InternVL3-1B-hf",
cache_dir="/media/jerem/641C8D6C1C8D3A56/hf_cache",
fps=1, # Frames per second to analyze
action_delay=0.5): # Delay between actions
"""
Initialize streaming Jarvis
Args:
fps: How many frames per second to analyze (1-10 recommended)
action_delay: Minimum delay between actions in seconds
"""
self.model_name = "CFP-Jarvis1-Stream"
self.version = "1.0.0"
self.fps = min(fps, 10) # Cap at 10 FPS
self.action_delay = action_delay
# Streaming components
self.streaming = False
self.capture_thread = None
self.analysis_thread = None
self.action_thread = None
# Queues for pipeline
self.frame_queue = queue.Queue(maxsize=10)
self.action_queue = queue.Queue(maxsize=100)
# Model components
self.processor = None
self.model = None
self.device = None
self.dtype = None
# Screen capture
self.sct = mss.mss()
# Performance metrics
self.last_capture_time = 0
self.last_analysis_time = 0
self.frames_analyzed = 0
# Setup cache
os.environ['HF_HOME'] = cache_dir
self.model_path = model_path
print(f"🎬 {self.model_name} v{self.version}")
print(f"📹 Streaming at {self.fps} FPS")
def load_model(self):
"""Load the model for streaming"""
try:
print(f"📥 Loading model for streaming...")
self.processor = AutoProcessor.from_pretrained(
self.model_path,
trust_remote_code=True
)
if torch.cuda.is_available():
self.device = "cuda"
self.dtype = torch.bfloat16
print("🚀 GPU streaming enabled")
else:
self.device = "cpu"
self.dtype = torch.float32
print("💻 CPU streaming (slower)")
self.model = AutoModelForImageTextToText.from_pretrained(
self.model_path,
torch_dtype=self.dtype,
device_map="auto" if self.device == "cuda" else self.device,
trust_remote_code=True,
low_cpu_mem_usage=True
).eval()
# Compile model for faster inference (PyTorch 2.0+)
if hasattr(torch, 'compile') and self.device == "cuda":
print("⚡ Compiling model for faster streaming...")
self.model = torch.compile(self.model, mode="reduce-overhead")
print("✅ Model ready for streaming")
return True
except Exception as e:
print(f"❌ Failed to load model: {e}")
return False
def capture_screen(self):
"""Continuously capture screen frames"""
print("📹 Starting screen capture...")
while self.streaming:
try:
# Capture screen
screenshot = self.sct.grab(self.sct.monitors[0])
# Convert to PIL Image
img = Image.frombytes(
'RGB',
(screenshot.width, screenshot.height),
screenshot.bgra,
'raw',
'BGRX'
)
# Resize for faster processing
img.thumbnail((1280, 720), Image.Resampling.LANCZOS)
# Add to queue if not full
if not self.frame_queue.full():
self.frame_queue.put(img)
self.last_capture_time = time.time()
# Control FPS
time.sleep(1.0 / self.fps)
except Exception as e:
print(f"❌ Capture error: {e}")
time.sleep(1)
def analyze_frames(self, task="Monitor screen for actions"):
"""Analyze captured frames and generate actions"""
print("🔍 Starting frame analysis...")
while self.streaming:
try:
# Get frame from queue
if not self.frame_queue.empty():
img = self.frame_queue.get(timeout=1)
# Quick analysis prompt for streaming
prompt = f"Task: {task}\nWhat is the most important UI element visible? Provide coordinates if clickable."
# Generate response
start_time = time.time()
response = self._quick_analyze(img, prompt)
analysis_time = time.time() - start_time
self.frames_analyzed += 1
self.last_analysis_time = analysis_time
# Parse response for actions
action = self._parse_action(response)
if action:
self.action_queue.put(action)
# Show performance
if self.frames_analyzed % 10 == 0:
print(f"⚡ FPS: {1/analysis_time:.1f} | Frames: {self.frames_analyzed}")
else:
time.sleep(0.1)
except Exception as e:
print(f"❌ Analysis error: {e}")
time.sleep(1)
def _quick_analyze(self, image, prompt):
"""Quick analysis optimized for streaming"""
try:
messages = [{
"role": "user",
"content": [
{"type": "image", "image": image},
{"type": "text", "text": prompt}
]
}]
text = self.processor.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
inputs = self.processor(
text=text,
images=image,
return_tensors="pt"
)
inputs = {
k: v.to(self.device) if torch.is_tensor(v) else v
for k, v in inputs.items()
}
if 'pixel_values' in inputs:
inputs['pixel_values'] = inputs['pixel_values'].to(self.dtype)
# Fast generation with fewer tokens
with torch.no_grad():
outputs = self.model.generate(
**inputs,
max_new_tokens=50, # Short response for speed
do_sample=False,
num_beams=1 # Greedy for speed
)
response = self.processor.decode(
outputs[0][inputs['input_ids'].shape[1]:],
skip_special_tokens=True
)
return response
except Exception as e:
return f"Error: {str(e)}"
def _parse_action(self, response):
"""Parse response to extract actionable information"""
# Simple parsing - would need more sophisticated NLP in production
action = None
response_lower = response.lower()
if "click" in response_lower:
# Try to extract coordinates
import re
coords = re.findall(r'\((\d+),\s*(\d+)\)', response)
if coords:
x, y = int(coords[0][0]), int(coords[0][1])
action = Action(type="click", coordinates=(x, y), confidence=0.8)
elif "button" in response_lower or "clickable" in response_lower:
# Default to center if button mentioned
action = Action(type="click", coordinates=(640, 360), confidence=0.5)
return action
def execute_actions(self):
"""Execute queued actions"""
print("🎮 Starting action executor...")
while self.streaming:
try:
if not self.action_queue.empty():
action = self.action_queue.get(timeout=1)
if action.type == "click" and action.coordinates:
x, y = action.coordinates
print(f"🖱️ Clicking at ({x}, {y})")
pyautogui.click(x, y)
elif action.type == "type" and action.text:
print(f"⌨️ Typing: {action.text}")
pyautogui.typewrite(action.text)
# Delay between actions
time.sleep(self.action_delay)
else:
time.sleep(0.1)
except Exception as e:
print(f"❌ Action error: {e}")
time.sleep(1)
def start_streaming(self, task="Monitor and interact with screen", auto_execute=False):
"""
Start streaming analysis
Args:
task: Description of what to monitor/do
auto_execute: Whether to automatically execute detected actions
"""
if not self.model:
if not self.load_model():
return False
print(f"\n🎬 Starting {self.model_name} streaming mode")
print(f"📋 Task: {task}")
print(f"🤖 Auto-execute: {auto_execute}")
self.streaming = True
# Start capture thread
self.capture_thread = threading.Thread(target=self.capture_screen)
self.capture_thread.start()
# Start analysis thread
self.analysis_thread = threading.Thread(
target=self.analyze_frames,
args=(task,)
)
self.analysis_thread.start()
# Start action thread if auto-execute
if auto_execute:
self.action_thread = threading.Thread(target=self.execute_actions)
self.action_thread.start()
print("✅ Streaming started! Press Ctrl+C to stop.")
return True
def stop_streaming(self):
"""Stop streaming analysis"""
print("\n🛑 Stopping streaming...")
self.streaming = False
# Wait for threads to finish
if self.capture_thread:
self.capture_thread.join(timeout=2)
if self.analysis_thread:
self.analysis_thread.join(timeout=2)
if self.action_thread:
self.action_thread.join(timeout=2)
print(f"📊 Stats: {self.frames_analyzed} frames analyzed")
print("✅ Streaming stopped")
def get_status(self):
"""Get streaming status"""
return {
"streaming": self.streaming,
"frames_analyzed": self.frames_analyzed,
"queue_size": self.frame_queue.qsize(),
"actions_pending": self.action_queue.qsize(),
"avg_fps": 1/self.last_analysis_time if self.last_analysis_time > 0 else 0
}
# Example usage
if __name__ == "__main__":
import signal
jarvis = CFPJarvis1Stream(fps=2) # 2 FPS for testing
# Handle Ctrl+C
def signal_handler(sig, frame):
print("\n\n⚠️ Interrupt received")
jarvis.stop_streaming()
exit(0)
signal.signal(signal.SIGINT, signal_handler)
# Start streaming
jarvis.start_streaming(
task="Monitor screen and identify clickable buttons",
auto_execute=False # Don't auto-click for safety
)
# Keep running
try:
while jarvis.streaming:
time.sleep(1)
status = jarvis.get_status()
if status["frames_analyzed"] % 10 == 0 and status["frames_analyzed"] > 0:
print(f"📊 Status: {status}")
except KeyboardInterrupt:
pass |