alonsorobots/scrfd_320_batched

InsightFace Batch-Optimized Models (Max Batch 32)

Re-exported InsightFace models with proper dynamic batch support and no cross-frame contamination.

Version Comparison

Repository	Max Batch	Recommendation
This repo	1-32	✅ Recommended - Optimal performance
alonsorobots/scrfd_320_batched_64	1-64	For experimentation

Batch=32 is optimal. Testing on RTX 5090 shows batch=64 provides no additional throughput benefit.

Why These Models?

The original InsightFace ONNX models have issues with batch inference:

• buffalo_l detection model: hardcoded batch=1
• buffalo_l_batch detection model: broken - has cross-frame contamination due to reshape operations that flatten the batch dimension

These re-exports fix the dynamic_axes in the ONNX graph for true batch inference.

Models

Model	Task	Input Shape	Output	Batch	Speedup
scrfd_10g_320_batch.onnx	Face Detection	[N, 3, 320, 320]	boxes, landmarks	1-32	6×
arcface_w600k_r50_batch.onnx	Face Embedding	[N, 3, 112, 112]	512-dim vectors	1-32	10×

Performance (TensorRT FP16, RTX 5090)

SCRFD Face Detection

Batch Size	FPS	ms/frame
1	867	1.15
16	5,498	0.18

ArcFace Embeddings

Batch Size	FPS	ms/embedding
1	292	3.4
16	3,029	0.33

Usage

import numpy as np
import onnxruntime as ort

# Load model
sess = ort.InferenceSession("scrfd_10g_320_batch.onnx", 
                            providers=["TensorrtExecutionProvider", "CUDAExecutionProvider"])

# Batch inference
batch = np.random.randn(16, 3, 320, 320).astype(np.float32)
outputs = sess.run(None, {"input.1": batch})

# outputs[0-2]: scores per FPN level (stride 8, 16, 32)
# outputs[3-5]: bboxes per FPN level
# outputs[6-8]: keypoints per FPN level

Verified: No Batch Contamination

# Same frame processed alone vs in batch = identical results
single_output = sess.run(None, {"input.1": frame[np.newaxis, ...]})
batch[7] = frame
batch_output = sess.run(None, {"input.1": batch})

max_diff = np.max(np.abs(single_output[0] - batch_output[0][7]))
# max_diff < 1e-5 ✓

Re-export Process

These models were re-exported from InsightFace's PyTorch source using MMDetection with proper dynamic_axes:

dynamic_axes = {
    "input.1": {0: "batch"},
    "score_8": {0: "batch"},
    "score_16": {0: "batch"},
    # ... all outputs
}

See SCRFD_320_EXPORT_INSTRUCTIONS.md for details.

License

Non-commercial research purposes only - per InsightFace license.

For commercial licensing, contact: [email protected]

Credits

• Original models: InsightFace by Jia Guo et al.
• SCRFD paper: Sample and Computation Redistribution for Efficient Face Detection
• ArcFace paper: ArcFace: Additive Angular Margin Loss for Deep Face Recognition