Computer Vision & ML Engineer

Dreamvu; Hyderabad; On-site; INR 1,200,000 - 2,000,000; Full time

Job summary

Dreamvu in Hyderabad is seeking a hands-on computer vision/ML engineer with 2+ years of experience to build systems that run on real data. The role requires strong foundations in multi-view geometry, Python and PyTorch, and experience with C++ for performance-critical code.

Responsibilities include training and evaluating deep models end-to-end, and developing data pipelines while operating on Linux with multi-GPU setups and cloud storage. Publications and VR/robotics tooling are a plus.

Qualifications

• 2+ years in computer vision or ML building data-driven systems.
• BTech/MTech/MS/PhD in CS/EE or related field.
• Strong multi-view geometry fundamentals (camera models, calibration, stereo, epipolar geometry, triangulation, pose estimation).
• Proficient Python and PyTorch; working C++ for performance-heavy code.
• Experience training, fine-tuning and evaluating deep models end-to-end and building the data pipeline.
• Hands-on experience with 3D hand/body pose, depth estimation, 3D reconstruction, or video understanding.
• Comfort on Linux with multi-GPU training, large video datasets and cloud storage.
• Ability to verify results and document work for reproducibility.

Skills

Computer vision

Python

PyTorch

C++

3D vision

Multiview geometry

Linux

Data pipelines

Education

BTech/MTech/MS/PhD in CS/EE

Tools

ROS

Isaac Sim

MuJoCo

LeRobot

Job description

Location : Hyderabad 5 days work from office

• 2+ years of hands-on computer vision or machine learning work, in research or industry, building systems that run on real data.
• BTech, MTech, MS or PhD in computer science, electrical engineering or a related field.
• Strong multi-view geometry fundamentals camera models, calibration, stereo, epipolar geometry, triangulation and pose estimation.
• Strong Python and PyTorch; working C++ for performance-critical or geometry-heavy code.
• Experience training, fine-tuning and evaluating deep models end to end, including building the data pipeline that feeds them.
• Hands-on experience with at least one of: 3D hand or body pose, depth estimation, 3D reconstruction, or video understanding.
• Comfort on Linux with multi-GPU training, large video datasets and cloud object storage.
• The discipline to verify results before reporting them, and to document your work clearly enough that someone else can reproduce it.

Strongly Preferred

• Publications at CVPR, ICCV, ECCV, NeurIPS, ICLR, ICRA, CoRL, 3DV or similar venues.
• 3D Gaussian splatting, NeRFs or classical multi-view stereo in practice.
• Fine-tuning VLMs or VLAs LoRA, SFT, instruction data design and evaluation.
• Foundation segmentation and detection models (SAM-family, open-vocabulary detectors) in automated labelling pipelines.
• Robotics tooling - ROS, Isaac Sim, MuJoCo, LeRobot - and an understanding of how robot learning consumes data.
• Wide-FOV, fisheye or omnidirectional camera models.
• Egocentric datasets and the specific failure modes of head- and wrist-mounted capture.