senior-computer-vision（高级计算机视觉）

Senior Computer Vision Engineer

The agent designs end-to-end computer vision pipelines for object detection, instance/semantic segmentation, and production deployment. It generates training configurations for YOLO/Detectron2/MMDetection, optimizes models for ONNX/TensorRT/OpenVINO runtimes, and builds dataset preparation workflows with format conversion and augmentation.

Quick Start

# Generate training configuration for YOLO or Faster R-CNN
python scripts/vision_model_trainer.py models/ --task detection --arch yolov8

# Analyze model for optimization opportunities (quantization, pruning)
python scripts/inference_optimizer.py model.pt --target onnx --benchmark

# Build dataset pipeline with augmentations
python scripts/dataset_pipeline_builder.py images/ --format coco --augment

Workflow 1: Object Detection Pipeline

The agent uses this workflow when building an object detection system from scratch.

Step 1: Define Detection Requirements

Analyze the detection task requirements:

Detection Requirements Analysis:
- Target objects: [list specific classes to detect]
- Real-time requirement: [yes/no, target FPS]
- Accuracy priority: [speed vs accuracy trade-off]
- Deployment target: [cloud GPU, edge device, mobile]
- Dataset size: [number of images, annotations per class]

Step 2: Select Detection Architecture

Choose architecture based on requirements:

Step 3: Prepare Dataset

Convert annotations to required format:

# COCO format (recommended)
python scripts/dataset_pipeline_builder.py data/images/ \
    --annotations data/labels/ \
    --format coco \
    --split 0.8 0.1 0.1 \
    --output data/coco/

# Verify dataset
python -c "from pycocotools.coco import COCO; coco = COCO('data/coco/train.json'); print(f'Images: {len(coco.imgs)}, Categories: {len(coco.cats)}')"

Step 4: Configure Training

Generate training configuration:

# For Ultralytics YOLO
python scripts/vision_model_trainer.py data/coco/ \
    --task detection \
    --arch yolov8m \
    --epochs 100 \
    --batch 16 \
    --imgsz 640 \
    --output configs/

# For Detectron2
python scripts/vision_model_trainer.py data/coco/ \
    --task detection \
    --arch faster_rcnn_R_50_FPN \
    --framework detectron2 \
    --output configs/

Step 5: Train and Validate

# Ultralytics training
yolo detect train data=data.yaml model=yolov8m.pt epochs=100 imgsz=640

# Detectron2 training
python train_net.py --config-file configs/faster_rcnn.yaml --num-gpus 1

# Validate on test set
yolo detect val model=runs/detect/train/weights/best.pt data=data.yaml

Step 6: Evaluate Results

Key metrics to analyze:

Workflow 2: Model Optimization and Deployment

Use this workflow when preparing a trained model for production deployment.

Step 1: Benchmark Baseline Performance

# Measure current model performance
python scripts/inference_optimizer.py model.pt \
    --benchmark \
    --input-size 640 640 \
    --batch-sizes 1 4 8 16 \
    --warmup 10 \
    --iterations 100

Expected output:

Baseline Performance (PyTorch FP32):
- Batch 1: 45.2ms (22.1 FPS)
- Batch 4: 89.4ms (44.7 FPS)
- Batch 8: 165.3ms (48.4 FPS)
- Memory: 2.1 GB
- Parameters: 25.9M

Step 2: Select Optimization Strategy

Step 3: Export to ONNX

# Export with dynamic batch size
python scripts/inference_optimizer.py model.pt \
    --export onnx \
    --input-size 640 640 \
    --dynamic-batch \
    --simplify \
    --output model.onnx

# Verify ONNX model
python -c "import onnx; model = onnx.load('model.onnx'); onnx.checker.check_model(model); print('ONNX model valid')"

Step 4: Apply Quantization (Optional)

For INT8 quantization with calibration:

# Generate calibration dataset
python scripts/inference_optimizer.py model.onnx \
    --quantize int8 \
    --calibration-data data/calibration/ \
    --calibration-samples 500 \
    --output model_int8.onnx

Quantization impact analysis:

Step 5: Convert to Target Runtime

# TensorRT (NVIDIA GPU)
trtexec --onnx=model.onnx --saveEngine=model.engine --fp16

# OpenVINO (Intel)
mo --input_model model.onnx --output_dir openvino/

# CoreML (Apple)
python -c "import coremltools as ct; model = ct.convert('model.onnx'); model.save('model.mlpackage')"

Step 6: Benchmark Optimized Model

python scripts/inference_optimizer.py model.engine \
    --benchmark \
    --runtime tensorrt \
    --compare model.pt

Expected speedup:

Optimization Results:
- Original (PyTorch FP32): 45.2ms
- Optimized (TensorRT FP16): 12.8ms
- Speedup: 3.5x
- Accuracy change: -0.3% mAP

Workflow 3: Custom Dataset Preparation

Use this workflow when preparing a computer vision dataset for training.

Step 1: Audit Raw Data

# Analyze image dataset
python scripts/dataset_pipeline_builder.py data/raw/ \
    --analyze \
    --output analysis/

Analysis report includes:

Dataset Analysis:
- Total images: 5,234
- Image sizes: 640x480 to 4096x3072 (variable)
- Formats: JPEG (4,891), PNG (343)
- Corrupted: 12 files
- Duplicates: 45 pairs

Annotation Analysis:
- Format detected: Pascal VOC XML
- Total annotations: 28,456
- Classes: 5 (car, person, bicycle, dog, cat)
- Distribution: car (12,340), person (8,234), bicycle (3,456), dog (2,890), cat (1,536)
- Empty images: 234

Step 2: Clean and Validate

# Remove corrupted and duplicate images
python scripts/dataset_pipeline_builder.py data/raw/ \
    --clean \
    --remove-corrupted \
    --remove-duplicates \
    --output data/cleaned/

Step 3: Convert Annotation Format

# Convert VOC to COCO format
python scripts/dataset_pipeline_builder.py data/cleaned/ \
    --annotations data/annotations/ \
    --input-format voc \
    --output-format coco \
    --output data/coco/

Supported format conversions:

Step 4: Apply Augmentations

# Generate augmentation config
python scripts/dataset_pipeline_builder.py data/coco/ \
    --augment \
    --aug-config configs/augmentation.yaml \
    --output data/augmented/

Recommended augmentations for detection:

# configs/augmentation.yaml
augmentations:
  geometric:
    - horizontal_flip: { p: 0.5 }
    - vertical_flip: { p: 0.1 }  # Only if orientation invariant
    - rotate: { limit: 15, p: 0.3 }
    - scale: { scale_limit: 0.2, p: 0.5 }

  color:
    - brightness_contrast: { brightness_limit: 0.2, contrast_limit: 0.2, p: 0.5 }
    - hue_saturation: { hue_shift_limit: 20, sat_shift_limit: 30, p: 0.3 }
    - blur: { blur_limit: 3, p: 0.1 }

  advanced:
    - mosaic: { p: 0.5 }  # YOLO-style mosaic
    - mixup: { p: 0.1 }   # Image mixing
    - cutout: { num_holes: 8, max_h_size: 32, max_w_size: 32, p: 0.3 }

Step 5: Create Train/Val/Test Splits

python scripts/dataset_pipeline_builder.py data/augmented/ \
    --split 0.8 0.1 0.1 \
    --stratify \
    --seed 42 \
    --output data/final/

Split strategy guidelines:

Step 6: Generate Dataset Configuration

# For Ultralytics YOLO
python scripts/dataset_pipeline_builder.py data/final/ \
    --generate-config yolo \
    --output data.yaml

# For Detectron2
python scripts/dataset_pipeline_builder.py data/final/ \
    --generate-config detectron2 \
    --output detectron2_config.py

Architecture Selection Guide

Object Detection Architectures

Segmentation Architectures

CNN vs Vision Transformer Trade-offs

Reference Documentation

1. Computer Vision Architectures

See references/computer_vision_architectures.md for:

• CNN backbone architectures (ResNet, EfficientNet, ConvNeXt)
• Vision Transformer variants (ViT, DeiT, Swin)
• Detection heads (anchor-based vs anchor-free)
• Feature Pyramid Networks (FPN, BiFPN, PANet)
• Neck architectures for multi-scale detection

2. Object Detection Optimization

See references/object_detection_optimization.md for:

• Non-Maximum Suppression variants (NMS, Soft-NMS, DIoU-NMS)
• Anchor optimization and anchor-free alternatives
• Loss function design (focal loss, GIoU, CIoU, DIoU)
• Training strategies (warmup, cosine annealing, EMA)
• Data augmentation for detection (mosaic, mixup, copy-paste)

3. Production Vision Systems

See references/production_vision_systems.md for:

• ONNX export and optimization
• TensorRT deployment pipeline
• Batch inference optimization
• Edge device deployment (Jetson, Intel NCS)
• Model serving with Triton
• Video processing pipelines

Common Commands

Ultralytics YOLO

# Training
yolo detect train data=coco.yaml model=yolov8m.pt epochs=100 imgsz=640

# Validation
yolo detect val model=best.pt data=coco.yaml

# Inference
yolo detect predict model=best.pt source=images/ save=True

# Export
yolo export model=best.pt format=onnx simplify=True dynamic=True

Detectron2

# Training
python train_net.py --config-file configs/COCO-Detection/faster_rcnn_R_50_FPN_3x.yaml \
    --num-gpus 1 OUTPUT_DIR ./output

# Evaluation
python train_net.py --config-file configs/faster_rcnn.yaml --eval-only \
    MODEL.WEIGHTS output/model_final.pth

# Inference
python demo.py --config-file configs/faster_rcnn.yaml \
    --input images/*.jpg --output results/ \
    --opts MODEL.WEIGHTS output/model_final.pth

MMDetection

# Training
python tools/train.py configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py

# Testing
python tools/test.py configs/faster_rcnn.py checkpoints/latest.pth --eval bbox

# Inference
python demo/image_demo.py demo.jpg configs/faster_rcnn.py checkpoints/latest.pth

Model Optimization

# ONNX export and simplify
python -c "import torch; model = torch.load('model.pt'); torch.onnx.export(model, torch.randn(1,3,640,640), 'model.onnx', opset_version=17)"
python -m onnxsim model.onnx model_sim.onnx

# TensorRT conversion
trtexec --onnx=model.onnx --saveEngine=model.engine --fp16 --workspace=4096

# Benchmark
trtexec --loadEngine=model.engine --batch=1 --iterations=1000 --avgRuns=100

Performance Targets

Resources

• Architecture Guide: references/computer_vision_architectures.md
• Optimization Guide: references/object_detection_optimization.md
• Deployment Guide: references/production_vision_systems.md
• Scripts: scripts/ directory for automation tools

Anti-Patterns

• Training without data audit -- skipping dataset_pipeline_builder.py analyze leads to corrupted images, duplicate pairs, and class imbalance surprises mid-training
• Deploying FP32 to production -- always export to FP16 minimum; FP32 wastes 2x memory and 1.5-2x latency for <0.5% mAP difference
• Ignoring calibration dataset -- INT8 quantization with random samples causes 5-10% mAP drop; use 500+ representative images from the training distribution
• One-size-fits-all architecture -- using YOLOv8x for edge deployment or YOLOv8n for high-accuracy requirements; match architecture to deployment target
• Benchmarking without warmup -- first N inference calls include JIT compilation overhead; always use --warmup 10 for accurate measurements
• Skipping ONNX validation -- export can silently produce incorrect models; always run onnx.checker.check_model() after export

Troubleshooting

Success Criteria

• Detection accuracy: mAP@50 above 0.70 and mAP@50:95 above 0.50 on the target validation set
• Inference latency: P99 latency under 50ms per frame at batch size 1 on target hardware for real-time deployments
• Throughput: Sustained processing above 30 FPS for real-time pipelines, above 10 FPS for high-accuracy pipelines
• Model size: Optimized model under 50MB for edge deployment, under 200MB for cloud GPU deployment
• Quantization fidelity: Less than 2% mAP drop when moving from FP32 to FP16; less than 3% drop for INT8
• Dataset quality: Class imbalance ratio no worse than 1:10 between least and most frequent classes; zero corrupted images; annotation coverage above 95% of images
• Deployment reliability: ONNX model passes onnx.checker.check_model() validation; TensorRT engine builds without warnings on target GPU architecture

Scope & Limitations

This skill covers:

• End-to-end object detection and segmentation pipeline design (data preparation through production deployment)
• Training configuration generation for Ultralytics YOLO, Detectron2, and MMDetection frameworks
• Model optimization and export to ONNX, TensorRT, OpenVINO, and CoreML runtimes
• Dataset format conversion (COCO, YOLO, Pascal VOC, CVAT), splitting, validation, and augmentation configuration

This skill does NOT cover:

• Generative vision tasks (image generation, style transfer, super-resolution) -- see dedicated generative AI skills
• 3D reconstruction, SLAM, or point cloud processing beyond basic depth estimation
• Medical imaging regulatory compliance (DICOM, FDA 510(k)) -- see ra-qm-team/ compliance skills
• Real-time video streaming infrastructure (RTSP, WebRTC, GStreamer pipeline design) -- see senior-devops for infrastructure

Integration Points

Tool Reference

vision_model_trainer.py

Purpose: Generates training configuration files for object detection and segmentation models across Ultralytics YOLO, Detectron2, and MMDetection frameworks.

Usage:

python scripts/vision_model_trainer.py <data_dir> [options]

Parameters:

Example:

# Generate Ultralytics YOLO training config
python scripts/vision_model_trainer.py data/coco/ --task detection --arch yolov8m --epochs 100 --batch 16 --output configs/train.yaml

# Analyze dataset only
python scripts/vision_model_trainer.py data/coco/ --analyze-only --json

# Generate Detectron2 config
python scripts/vision_model_trainer.py data/coco/ --framework detectron2 --arch faster_rcnn_R_50_FPN --output configs/detectron2.py

Output Formats:

• Human-readable (default): Prints a summary table with framework, architecture, parameters, COCO mAP, and the training command
• JSON (--json): Full configuration dictionary including all hyperparameters and metadata
• Config file (--output): YAML for Ultralytics; Python config for Detectron2/MMDetection

inference_optimizer.py

Purpose: Analyzes model structure, benchmarks inference speed across batch sizes, and provides optimization recommendations for target deployment platforms.

Usage:

python scripts/inference_optimizer.py <model_path> [options]

Parameters:

Example:

# Analyze model structure
python scripts/inference_optimizer.py model.onnx --analyze

# Benchmark with custom batch sizes
python scripts/inference_optimizer.py model.pt --benchmark --input-size 640 640 --batch-sizes 1 4 8 16 --warmup 10 --iterations 100

# Get optimization recommendations for edge deployment
python scripts/inference_optimizer.py model.pt --analyze --recommend --target edge --json

# Save full report
python scripts/inference_optimizer.py model.onnx --analyze --benchmark --recommend --output report.json

Output Formats:

• Human-readable (default): Summary table with file size, parameters, node count; benchmark table with latency, throughput, and P99 per batch size; numbered optimization recommendations with expected speedup
• JSON (--json): Nested dictionary with analysis, benchmark, and recommendations keys
• File (--output): JSON report saved to specified path

dataset_pipeline_builder.py

Purpose: Production-grade tool for analyzing, converting, splitting, augmenting, and validating computer vision datasets. Uses subcommands for each operation.

Usage:

python scripts/dataset_pipeline_builder.py <command> [options]

Subcommands:

analyze -- Analyze dataset structure and statistics

python scripts/dataset_pipeline_builder.py analyze --input data/coco/
python scripts/dataset_pipeline_builder.py analyze --input data/coco/ --json

convert -- Convert between annotation formats

python scripts/dataset_pipeline_builder.py convert --input data/voc/ --output data/coco/ --format coco
python scripts/dataset_pipeline_builder.py convert --input data/coco/ --output data/yolo/ --format yolo --source-format coco

split -- Split dataset into train/val/test sets

python scripts/dataset_pipeline_builder.py split --input data/coco/ --train 0.8 --val 0.1 --test 0.1 --stratify --seed 42

augment-config -- Generate augmentation configuration

python scripts/dataset_pipeline_builder.py augment-config --task detection --intensity heavy --output augmentations.yaml

validate -- Validate dataset integrity

python scripts/dataset_pipeline_builder.py validate --input data/coco/ --format coco

Output Formats:

• Human-readable (default): Structured report with dataset statistics, annotation counts, class distributions, quality checks, and actionable recommendations
• JSON (--json): Full analysis dictionary including image stats, annotation details, bounding box statistics, and quality check results