Spaces:
Running
Running
File size: 9,504 Bytes
aa8e154 | 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 | """
modules/expression_detection.py
ML-based expression detection using trained FER2013 CNN.
Falls back to rule-based if model not found.
"""
import cv2
import numpy as np
import os
import json
from collections import deque
# ── Model path (relative to project root) ────────────────────────────────────
_MODEL_DIR = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "models")
_MODEL_PATH = os.path.join(_MODEL_DIR, "best_expression_model.keras")
_CMAP_PATH = os.path.join(_MODEL_DIR, "expression_class_map.json")
# Nervousness weights per expression (higher = more nervous)
_NERVOUSNESS_WEIGHT = {
"Angry": 80,
"Disgust": 70,
"Fear": 90,
"Happy": 10,
"Neutral": 20,
"Sad": 60,
"Surprise": 40,
}
# Positive expression score per class (higher = more confident-looking)
_EXPR_SCORE = {
"Angry": 30,
"Disgust": 20,
"Fear": 10,
"Happy": 95,
"Neutral": 70,
"Sad": 25,
"Surprise": 55,
}
class ExpressionDetector:
def __init__(self, fps: int = 20):
self._fps = fps
self._model = None
self._class_map = None # {0: "Angry", 1: "Disgust", ...}
self._ml_ready = False
# Blink tracking (kept for blink_rate metric)
self._blink_buffer = deque(maxlen=fps * 10)
self._blink_count = 0
self._blink_frames = 0
self._EAR_THRESHOLD = 0.22
self._CONSEC_FRAMES = 2
# Smoothing buffer for expression predictions
self._expr_buffer = deque(maxlen=8)
self._load_model()
def _load_model(self):
if not os.path.exists(_MODEL_PATH):
print(f"[ExpressionDetector] Model not found at {_MODEL_PATH}. Using rule-based fallback.")
return
try:
import tensorflow as tf
self._model = tf.keras.models.load_model(_MODEL_PATH)
if os.path.exists(_CMAP_PATH):
with open(_CMAP_PATH) as f:
raw = json.load(f)
self._class_map = {int(k): v for k, v in raw.items()}
else:
# default FER2013 order
self._class_map = {
0: "Angry", 1: "Disgust", 2: "Fear",
3: "Happy", 4: "Neutral", 5: "Sad", 6: "Surprise",
}
self._ml_ready = True
print("[ExpressionDetector] ML model loaded successfully.")
except Exception as e:
print(f"[ExpressionDetector] Model load failed: {e}. Using rule-based fallback.")
# ── Public API ────────────────────────────────────────────────────────────
def detect(self, key_points: dict, frame_shape: tuple, frame_bgr=None) -> dict:
"""
Returns:
expression : str
expression_score : int 0-100
nervousness_score: int 0-100
blink_rate : int (blinks/min)
ml_confidence : float (model softmax confidence, 0 if rule-based)
"""
blink_rate = self._update_blink(key_points, frame_shape)
if self._ml_ready and frame_bgr is not None:
result = self._ml_detect(key_points, frame_bgr)
else:
result = self._rule_based_detect(key_points, frame_shape)
result["blink_rate"] = blink_rate
return result
# ── ML Detection ──────────────────────────────────────────────────────────
def _ml_detect(self, key_points: dict, frame_bgr: np.ndarray) -> dict:
face_roi = self._extract_face_roi(key_points, frame_bgr)
if face_roi is None:
return self._fallback_neutral()
try:
gray = cv2.cvtColor(face_roi, cv2.COLOR_BGR2GRAY)
resized = cv2.resize(gray, (48, 48))
inp = resized.astype(np.float32) / 255.0
inp = inp.reshape(1, 48, 48, 1)
probs = self._model.predict(inp, verbose=0)[0]
pred_idx = int(np.argmax(probs))
ml_conf = float(probs[pred_idx])
# Smoothing — majority vote over last N frames
self._expr_buffer.append(pred_idx)
from collections import Counter
smoothed_idx = Counter(self._expr_buffer).most_common(1)[0][0]
expression = self._class_map.get(smoothed_idx, "Neutral")
expr_score = _EXPR_SCORE.get(expression, 50)
nerv_score = _NERVOUSNESS_WEIGHT.get(expression, 30)
return {
"expression": expression,
"expression_score": expr_score,
"nervousness_score": nerv_score,
"ml_confidence": round(ml_conf, 3),
}
except Exception as e:
print(f"[ExpressionDetector] ML inference error: {e}")
return self._fallback_neutral()
def _extract_face_roi(self, key_points: dict, frame_bgr: np.ndarray):
"""Crop face region using nose + eye landmarks."""
try:
h, w = frame_bgr.shape[:2]
nose = key_points.get("nose_tip")
l_eye = key_points.get("left_eye")
r_eye = key_points.get("right_eye")
if nose is None or l_eye is None or r_eye is None:
return None
cx = int(nose[0] * w)
cy = int(nose[1] * h)
eye_dist = abs(int(l_eye[0] * w) - int(r_eye[0] * w))
pad = max(eye_dist, 60)
x1 = max(0, cx - pad)
y1 = max(0, cy - pad)
x2 = min(w, cx + pad)
y2 = min(h, cy + pad)
roi = frame_bgr[y1:y2, x1:x2]
return roi if roi.size > 0 else None
except Exception:
return None
# ── Rule-based Fallback ───────────────────────────────────────────────────
def _rule_based_detect(self, key_points: dict, frame_shape: tuple) -> dict:
mouth_open = self._mouth_open_ratio(key_points, frame_shape)
eyebrow_raise= self._eyebrow_raise(key_points, frame_shape)
if mouth_open > 0.05 and eyebrow_raise > 0.03:
expression = "Surprise"
elif mouth_open > 0.04:
expression = "Happy"
elif eyebrow_raise < -0.01:
expression = "Angry"
else:
expression = "Neutral"
return {
"expression": expression,
"expression_score": _EXPR_SCORE.get(expression, 50),
"nervousness_score": _NERVOUSNESS_WEIGHT.get(expression, 30),
"ml_confidence": 0.0,
}
def _fallback_neutral(self) -> dict:
return {
"expression": "Neutral",
"expression_score": 70,
"nervousness_score": 20,
"ml_confidence": 0.0,
}
# ── Blink Rate ────────────────────────────────────────────────────────────
def _update_blink(self, key_points: dict, frame_shape: tuple) -> int:
ear = self._eye_aspect_ratio(key_points, frame_shape)
if ear < self._EAR_THRESHOLD:
self._blink_frames += 1
else:
if self._blink_frames >= self._CONSEC_FRAMES:
self._blink_count += 1
self._blink_frames = 0
self._blink_buffer.append(1)
window_sec = len(self._blink_buffer) / self._fps
if window_sec > 0:
return int(self._blink_count / window_sec * 60)
return 0
def _eye_aspect_ratio(self, key_points: dict, frame_shape: tuple) -> float:
try:
h, w = frame_shape[:2]
le = key_points.get("left_eye")
re = key_points.get("right_eye")
if le is None or re is None:
return 0.3
eye_w = abs(le[0] - re[0]) * w
eye_h = max(le[1], re[1]) * h * 0.15
return eye_h / (eye_w + 1e-6)
except Exception:
return 0.3
def _mouth_open_ratio(self, key_points: dict, frame_shape: tuple) -> float:
try:
h, w = frame_shape[:2]
top = key_points.get("upper_lip")
bot = key_points.get("lower_lip")
if top is None or bot is None:
return 0.0
return abs(top[1] - bot[1])
except Exception:
return 0.0
def _eyebrow_raise(self, key_points: dict, frame_shape: tuple) -> float:
try:
nose = key_points.get("nose_tip")
leb = key_points.get("left_eyebrow")
reb = key_points.get("right_eyebrow")
if nose is None or leb is None or reb is None:
return 0.0
avg_brow = (leb[1] + reb[1]) / 2
return nose[1] - avg_brow
except Exception:
return 0.0 |