Spaces:
Sleeping
Sleeping
Update models.py
Browse files
models.py
CHANGED
|
@@ -1,11 +1,12 @@
|
|
| 1 |
# ==============================
|
| 2 |
-
# MODEL +
|
| 3 |
# ==============================
|
| 4 |
|
| 5 |
from transformers import pipeline
|
| 6 |
import torch
|
| 7 |
import time
|
| 8 |
from datetime import datetime
|
|
|
|
| 9 |
|
| 10 |
device = 0 if torch.cuda.is_available() else -1
|
| 11 |
|
|
@@ -126,58 +127,13 @@ def multimodal_analyze(text, image, audio):
|
|
| 126 |
"""
|
| 127 |
except Exception as e:
|
| 128 |
audio_result_display = f"Audio processing error: {str(e)}"
|
| 129 |
-
audio_label = None
|
| 130 |
-
audio_conf = 0
|
| 131 |
|
| 132 |
# -------- FUSION --------
|
| 133 |
-
reasoning_lines =
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
reasoning_lines.append(
|
| 139 |
-
f"Text suggests a {text_label.lower()} tone ({text_conf}%)."
|
| 140 |
-
)
|
| 141 |
-
weight = 0.4
|
| 142 |
-
fusion_score += text_conf * weight if text_label == "POSITIVE" else -text_conf * weight
|
| 143 |
-
total_weight += weight
|
| 144 |
-
|
| 145 |
-
if audio_label:
|
| 146 |
-
reasoning_lines.append(
|
| 147 |
-
f"Spoken content carries a {audio_label.lower()} tone ({audio_conf}%)."
|
| 148 |
-
)
|
| 149 |
-
weight = 0.35
|
| 150 |
-
fusion_score += audio_conf * weight if audio_label == "POSITIVE" else -audio_conf * weight
|
| 151 |
-
total_weight += weight
|
| 152 |
-
|
| 153 |
-
if image_label and image_conf > 40:
|
| 154 |
-
reasoning_lines.append(
|
| 155 |
-
f"Image classified as {image_label} ({image_conf}%)."
|
| 156 |
-
)
|
| 157 |
-
weight = 0.25
|
| 158 |
-
fusion_score += image_conf * weight
|
| 159 |
-
total_weight += weight
|
| 160 |
-
|
| 161 |
-
if total_weight > 0:
|
| 162 |
-
fusion_score = fusion_score / total_weight
|
| 163 |
-
|
| 164 |
-
fusion_score = max(min(fusion_score, 100), -100)
|
| 165 |
-
|
| 166 |
-
if fusion_score > 60:
|
| 167 |
-
alignment_message = "Strong positive multimodal alignment detected."
|
| 168 |
-
color = "#22c55e"
|
| 169 |
-
elif fusion_score > 20:
|
| 170 |
-
alignment_message = "Moderate positive contextual signals observed."
|
| 171 |
-
color = "#facc15"
|
| 172 |
-
elif fusion_score > -20:
|
| 173 |
-
alignment_message = "Mixed or neutral signals across modalities."
|
| 174 |
-
color = "#94a3b8"
|
| 175 |
-
elif fusion_score > -60:
|
| 176 |
-
alignment_message = "Moderate negative contextual alignment detected."
|
| 177 |
-
color = "#f97316"
|
| 178 |
-
else:
|
| 179 |
-
alignment_message = "Strong negative multimodal alignment detected."
|
| 180 |
-
color = "#ef4444"
|
| 181 |
|
| 182 |
processing_time = round(time.time() - start_time, 2)
|
| 183 |
|
|
|
|
| 1 |
# ==============================
|
| 2 |
+
# MODEL + MODALITY PROCESSING
|
| 3 |
# ==============================
|
| 4 |
|
| 5 |
from transformers import pipeline
|
| 6 |
import torch
|
| 7 |
import time
|
| 8 |
from datetime import datetime
|
| 9 |
+
from fusion import compute_fusion
|
| 10 |
|
| 11 |
device = 0 if torch.cuda.is_available() else -1
|
| 12 |
|
|
|
|
| 127 |
"""
|
| 128 |
except Exception as e:
|
| 129 |
audio_result_display = f"Audio processing error: {str(e)}"
|
|
|
|
|
|
|
| 130 |
|
| 131 |
# -------- FUSION --------
|
| 132 |
+
fusion_score, reasoning_lines, alignment_message, color = compute_fusion(
|
| 133 |
+
text_label, text_conf,
|
| 134 |
+
image_label, image_conf,
|
| 135 |
+
audio_label, audio_conf
|
| 136 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
processing_time = round(time.time() - start_time, 2)
|
| 139 |
|