fix bug

check_model.py

@@ -0,0 +1,59 @@
+#!/usr/bin/env python3
+"""
+Script to verify if the model has been trained with actual weights.
+"""
+import torch
+from ultralytics import YOLO
+from huggingface_hub import hf_hub_download
+
+# Download model
+repo_id = 'VietCat/GTSRB-Model'
+file_path = 'models/GTSRB.pt'
+local_model_path = hf_hub_download(repo_id=repo_id, filename=file_path)
+
+print(f"Model path: {local_model_path}\n")
+
+# Load model
+model = YOLO(local_model_path)
+
+print("="*80)
+print("MODEL WEIGHTS ANALYSIS")
+print("="*80)
+
+# Check model layers and weights
+print("\nChecking model weights...")
+total_params = 0
+zero_params = 0
+trained_params = 0
+
+for name, param in model.model.named_parameters():
+    param_count = param.numel()
+    total_params += param_count
+    
+    # Check if weights are mostly zeros or random initialization
+    if torch.allclose(param, torch.zeros_like(param), atol=1e-6):
+        zero_params += param_count
+        status = "ZERO"
+    elif param.mean().item() != 0:
+        trained_params += param_count
+        status = "TRAINED"
+    else:
+        status = "RANDOM"
+    
+    if param_count > 1000:  # Only print large layers
+        print(f"{name:50s} | {param_count:10,} params | {status:10s} | mean: {param.mean().item():.6f}, std: {param.std().item():.6f}")
+
+print(f"\n{'='*80}")
+print(f"Total parameters: {total_params:,}")
+print(f"Zero parameters: {zero_params:,} ({100*zero_params/total_params:.1f}%)")
+print(f"Trained parameters: {trained_params:,} ({100*trained_params/total_params:.1f}%)")
+
+# Check if this looks like trained weights
+if zero_params / total_params > 0.5:
+    print("\n⚠️  WARNING: Model has >50% zero parameters - may not be properly trained!")
+elif trained_params / total_params > 0.7:
+    print("\n✅ Model appears to be properly trained")
+else:
+    print("\n❓ Uncertain - model may need verification")
+
+print("="*80)

config.yaml

@@ -1,6 +1,6 @@
 model:
   path: 'VietCat/GTSRB-Model/models/GTSRB.pt' # Path to the YOLO model on Hugging Face Hub (will be downloaded automatically)
-  confidence_threshold: 0.25 # Minimum confidence for detections
+  confidence_threshold: 0.1 # Minimum confidence for detections
 
 inference:
   box_color: (128, 0, 128) # Purple color for bounding boxes (BGR format)