Dev/add benchmark

.gitignore

@@ -1,4 +1,5 @@
 data/
+output/
 checkpoints/
 
 # Byte-compiled / optimized / DLL files

demo.py

@@ -37,6 +37,8 @@
     else:
         weights_path = "naver/" + args.model_name
     model = AsymmetricCroCo3DStereo.from_pretrained(weights_path).to(args.device)
+
+    print(f"Hosting on {server_name}:{args.server_port}")
 
     # dust3r will write the 3D model inside tmpdirname
     with tempfile.TemporaryDirectory(suffix='dust3r_gradio_demo') as tmpdirname:

docker/docker-compose-cuda.yml

@@ -9,8 +9,12 @@ services:
     environment:
       - DEVICE=cuda
       - MODEL=${MODEL:-DUSt3R_ViTLarge_BaseDecoder_512_dpt.pth}
+      - PYTHONPATH=/dust3r  # <--- ADD THIS LINE
     volumes:
-      - ./files/checkpoints:/dust3r/checkpoints
+      - ./files/checkpoints:/checkpoints
+      - ./files/data:/data
+      - ./files/output:/output
+      - ../:/dust3r
     cap_add:
       - IPC_LOCK
       - SYS_RESOURCE

docker/files/cuda.Dockerfile

@@ -20,6 +20,8 @@ RUN pip install opencv-python==4.8.0.74
 WORKDIR /dust3r/croco/models/curope/
 RUN python setup.py build_ext --inplace
 
+RUN pip install boto3 zstandard
+
 WORKDIR /dust3r
 COPY entrypoint.sh /entrypoint.sh
 RUN chmod +x /entrypoint.sh

docker/files/entrypoint.sh

@@ -5,4 +5,5 @@ set -eux
 DEVICE=${DEVICE:-cuda}
 MODEL=${MODEL:-DUSt3R_ViTLarge_BaseDecoder_512_dpt.pth}
 
-exec python3 demo.py --weights "checkpoints/$MODEL" --device "$DEVICE" --local_network "$@"
+# Keep the container running for debugging
+tail -f /dev/null

docker/run.sh

@@ -35,10 +35,12 @@ set_dcomp() {
 run_docker() {
     export MODEL=${model_name}
     if [ "$with_cuda" -eq 1 ]; then
-        $dcomp -f docker-compose-cuda.yml up --build
+        $dcomp -f docker-compose-cuda.yml up --build -d
     else
-        $dcomp -f docker-compose-cpu.yml up --build
+        $dcomp -f docker-compose-cpu.yml up --build -d
     fi
+    echo "Docker container started in detached mode."
+    echo "To attach to the container, run: $dcomp exec dust3r-demo /bin/bash"
 }
 
 with_cuda=0

dust3r/analysis_scale.py

@@ -0,0 +1,67 @@
+import os
+import pandas as pd
+import matplotlib.pyplot as plt
+import argparse
+
+def plot_scale_statistics(results_path):
+    """Plot scale statistics from benchmark results.
+
+    Args:
+        results_path: Path to the CSV file containing benchmark results
+    """
+    # Read the results
+    df = pd.read_csv(results_path)
+
+    # Create figure with subplots
+    fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(12, 10))
+
+    # Plot 1: Scale factors over items
+    ax1.plot(df['item_id'], df['optimal_scale'], 'b.', alpha=0.5, label='Scale factors')
+    ax1.axhline(y=df['optimal_scale'].mean(), color='r', linestyle='--', label=f'Mean: {df["optimal_scale"].mean():.2f}')
+    ax1.set_xlabel('Item ID')
+    ax1.set_ylabel('Optimal Scale')
+    ax1.set_title('Scale Factors Distribution')
+    ax1.legend()
+    ax1.grid(True)
+
+    # Plot 2: Histogram of scale factors
+    ax2.hist(df['optimal_scale'], bins=30, alpha=0.7, color='b')
+    ax2.axvline(x=df['optimal_scale'].mean(), color='r', linestyle='--', 
+                label=f'Mean: {df["optimal_scale"].mean():.2f}')
+    ax2.axvline(x=df['optimal_scale'].median(), color='g', linestyle='--', 
+                label=f'Median: {df["optimal_scale"].median():.2f}')
+    ax2.set_xlabel('Scale Factor')
+    ax2.set_ylabel('Frequency')
+    ax2.set_title('Histogram of Scale Factors')
+    ax2.legend()
+    ax2.grid(True)
+
+    plt.tight_layout()
+    output_path = os.path.join(os.path.dirname(results_path), 'scale_statistics.png')
+    plt.savefig(output_path)
+    plt.close()
+
+    # Print statistics
+    print("\nScale Factor Statistics:")
+    print(f"Mean: {df['optimal_scale'].mean():.3f}")
+    print(f"Median: {df['optimal_scale'].median():.3f}")
+    print(f"Std: {df['optimal_scale'].std():.3f}")
+    print(f"Min: {df['optimal_scale'].min():.3f}")
+    print(f"Max: {df['optimal_scale'].max():.3f}")
+    print(f"25th percentile: {df['optimal_scale'].quantile(0.25):.3f}")
+    print(f"75th percentile: {df['optimal_scale'].quantile(0.75):.3f}")
+
+def main():
+    parser = argparse.ArgumentParser(description="Analyze scale factors from DUSt3R benchmark results.")
+    parser.add_argument('--results', type=str, required=True, help='Path to the benchmark results CSV file.')
+
+    args = parser.parse_args()
+
+    if not os.path.exists(args.results):
+        print(f"Results file not found at {args.results}")
+        return
+
+    plot_scale_statistics(args.results)
+
+if __name__ == "__main__":
+    main()