Azilen TechnologiesAI/ML - Computer Vision
AI/ML - Computer Vision
Job Description
Job Description
AI/ML Engineer – Computer Vision
1. Key Responsibilities
Dataset & Annotation
- Define annotation standards (bounding boxes, oriented bounding boxes, polygons, instance masks) based on the problem statement.
- Ensure high-quality annotations, dataset consistency, class balance, and adequate edge-case coverage.
- Manage dataset curation, validation, augmentation, and versioning for model training.
Model Development & Training
- Benchmark and evaluate different model architectures (YOLOv8/YOLOv11, YOLOv11-OBB, Instance Segmentation, Vision Transformers, etc.) based on accuracy, latency, and deployment requirements.
- Train, fine-tune, and optimize deep learning models using transfer learning, data augmentation, hyperparameter tuning, and hard-negative mining.
- Continuously improve model performance through experimentation and evaluation.
Computer Vision Development
- Develop image and video analytics pipelines using OpenCV and modern Computer Vision techniques.
- Implement object detection, object tracking, image segmentation, OCR, and feature extraction solutions as required.
- Develop robust preprocessing and post-processing pipelines for production-ready inference.
Model Deployment
- Export trained models to ONNX and optimize them for deployment using TensorRT, ONNX Runtime, or equivalent inference frameworks.
- Perform model optimization through quantization (INT8/PTQ), profiling, and performance tuning on GPU or edge devices.
- Develop scalable inference pipelines, including camera integration, preprocessing, inference, post-processing, and result generation.
Model Evaluation & MLOps
- Define and monitor model performance metrics such as mAP, Precision, Recall, F1-Score, FPS, and inference latency.
- Maintain reproducible training pipelines, dataset versioning, and model versioning.
- Support continuous model improvement through retraining and performance monitoring.
2. Mandatory Skills
|
Area |
Requirement |
|
Programming |
Strong proficiency in Python. Working knowledge of C++ for inference-side integration is preferred. |
|
Deep Learning Frameworks |
Hands-on experience with PyTorch. Experience with the Ultralytics YOLO ecosystem (training, fine-tuning, export, custom datasets). |
|
Computer Vision |
Strong understanding of Object Detection, Image Classification, Segmentation, Object Tracking, and Video Analytics. |
|
YOLO |
Practical experience with YOLO models (preferably YOLOv11), including training, evaluation, deployment, confidence threshold tuning, IoU, NMS, and performance optimization. |
|
OpenCV |
Strong experience with OpenCV for image processing, geometric transformations, camera integration, feature extraction, and video processing. |
|
Model Deployment |
Experience deploying AI models using ONNX, ONNX Runtime, TensorRT, or similar inference frameworks on GPU or edge devices. |
|
Model Optimization |
Hands-on experience with quantization, inference optimization, and performance tuning for production deployments. |
|
Data Pipeline |
Experience with annotation tools such as CVAT, Roboflow, or Label Studio. Knowledge of dataset preparation, augmentation, and data quality management. |
|
Engineering |
Git, Docker, Linux, virtual environments, modular coding practices, and API integration. |
|
Evaluation |
Strong understanding of model evaluation metrics including mAP, Precision, Recall, F1-Score, FPS, latency, and confusion matrix analysis. |
|
Communication |
Ability to explain technical decisions, model performance, and trade-offs to technical and non-technical stakeholders. |
3. Good to Have
|
Area |
Nice to Have |
|
Object Tracking |
Experience with ByteTrack, DeepSORT, BoT-SORT, or similar multi-object tracking algorithms. |
|
Oriented Detection |
Experience with YOLO-OBB, rotated bounding boxes, angle regression, and rotated-NMS. |
|
Segmentation |
Experience with Instance Segmentation models such as YOLO-Seg, Mask R-CNN, or SAM. |
|
OCR |
Experience with PaddleOCR, EasyOCR, Tesseract, or TrOCR. |
|
Vision Models |
Familiarity with Vision Transformers (ViT), CLIP, DINO, Grounding DINO, Florence-2, or similar modern vision models. |
|
Inference Runtime |
Experience with TensorRT, OpenVINO, ONNX Runtime, TFLite, or NVIDIA DeepStream. |
|
MLOps |
Experience with MLflow, Weights & Biases, DVC, model versioning, and automated training pipelines. |
|
Cloud & Deployment |
Exposure to AWS, Azure, GCP, Docker, Kubernetes, and scalable AI deployment. |
|
GPU Optimization |
Knowledge of CUDA, NVIDIA GPU optimization, mixed precision inference, and performance profiling. |
Quick Apply
~2 minApply through whichever channel suits you best.