Website:
zoidtech.co.in
Job details:
This role is NOT suitable for candidates whose experience is primarily in LLMs, generative AI, prompt engineering, RAG, agentic workflows, or API integration
Role Summary-
Role Name - Deep Learning Engineer – Vision
Experience- 1-3 years
Location - Sector 132, Noida
About ZOID
ZOID is a deep-tech R&D startup working on advanced technologies for the defence and aerospace sector. We build customised solutions for the Indian Military and UAV manufacturers, with our strength being hands-on expertise in AI, electronics, UAVs and robotics.
Current projects include:
- AI-based FOD detection software for naval air stations
- Reusable Off-board Missile Decoy for anti-ship missile defence
- GNSS-Denied Navigation Suite
- Aerial Intelligent Mapping Suite
- Vision-based Kinetic Strike Solution
- Swarming Solutions
Role Brief:
We are looking for an Deep Learning Engineer – Vision to handle computer-vision data operations through field data collection, image annotation, dataset QC, and test-set analysis for UAV/defence applications. The role involves working with field teams during runway/track tests and maintaining accurate, traceable datasets and test logs.
Key Responsibilities:
- Own the optics testing end to end (FOV, focal length and depth-of-field calculations) for detecting 3 x 3 x 3 mm objects.
- Plan and conduct optics testing including camera & lens settings (focus, motion-blur budget, exposure, gain, FPS, binning, etc).
- Build and manage the dataset, annotation and QA/QC rules, data versioning, and backup (Raid Configuration).
- Analyse the type of datasets which should be collected for any specific task.
- Develop the detection & classification model for millimetre-scale objects at high resolution.
- Optimise models for real-time inference on the GPU servers such that multiple cameras can simultaneously operate on a single GPU without any accuracy loss.
- Implement object localisation and geo-referencing of detections using RTK/GNSS and frame timestamps and validate CEP in the field.
- Maintain a model registry with regression tests so no release degrades detection performance; run the false-positive review loop with operations staff.
- Analyse why false positives are occurring and how to eliminate them?
- Recommend computation hardware (GPU, memory, storage, camera network cards) based on model and data-rate requirements.
- Apply AI methods to 3D photogrammetry data. Run segmentation, object detection and change detection, and extend the 3D processing pipeline where required. (Good to have)
Must Have:
- Fluency in Python.
- Strong computer vision fundamentals
- Knowledge of PyTorch & implementation of object detection at high resolution and small object scale (YOLO/DETR family or equivalent).
- Dataset engineering: annotation tooling (CVAT/Label Studio), QC methods, data versioning (DVC), experiment tracking (W&B/MLflow).
- TensorRT, CUDA, ONNX, INT8 quantisation, multi-GPU inference.
- Optics knowledge: FOV, IFOV, GSD, focal length, MTF, exposure & experience with machine-vision cameras and NIR imaging (Good to have)
- Object localisation and geo-referencing: pixel-to-ground mapping, CEP estimation, working with RTK/GNSS and time-synchronised frames. (Good to have)
- 3D data handling: point clouds and meshes (Open3D, PCL, PDAL or equivalent), knowledge of 3D deep learning models (PointNet/PointNet++, KPConv or similar) for segmentation and detection on 3D models. (Good to have)
- Linux, Git and CI; comfortable deploying and debugging on field hardware.
Good to Have:
- Experience with multi-camera systems: synchronisation, overlapping fields of view, and stitching or fusing outputs from a camera array.
- Familiarity with GigE-Vision/GenICam camera SDKs and PoE camera networks, enough to debug frame drops and timing with the software team.
- Active learning or semi-automatic labelling to cut annotation effort on large field datasets.
- Exposure to anomaly detection or unsupervised methods for finding objects that were never in the training set.
- LiDAR or depth-sensor data handling, useful for cross-checking optical detections and for AIMS 3D work.
- Understanding of EO/IR sensor behaviour beyond NIR (SWIR/MWIR), for future sensor upgrades.
- Basic Docker and model-serving experience for reproducible field deployments
Click on Apply to know more.
