Why computer vision is a strong specialization in India
Computer vision (CV) is one of the highest-leverage ML specializations for Indian engineers. Real Indian companies hire CV engineers for problems that genuinely matter:
Indian companies hiring CV engineers (2026):
Healthcare AI:
- Niramai (Bengaluru) — breast cancer detection via thermal imaging
- SigTuple (Bengaluru) — pathology image analysis
- Qure.ai (Mumbai) — radiology AI (chest X-rays, CT scans)
- HealthCubed (Bengaluru) — medical imaging
- Tricog (Bengaluru) — ECG analysis
Identity / KYC / Document AI:
- Hyperverge (Bengaluru) — face verification, ID document OCR (Aadhaar/PAN)
- IDfy (Mumbai) — KYC verification
- Signzy (Bengaluru) — onboarding AI
- Veri5Digital — KYC
Retail / E-commerce:
- Flipkart — visual search, product matching, return fraud detection
- Myntra — fashion AI, sizing, visual recommendations
- Tata Cliq — visual catalog
- Reliance Retail — store analytics
Autonomous / ADAS:
- Bosch India — ADAS development
- Continental India — vehicle vision
- Mercedes Benz R&D — autonomous driving
- Ola Electric — vehicle vision (limited)
- Mahindra Racing — vehicle vision
Geospatial / Satellite:
- ISRO — satellite image analysis
- Pixxel (Bengaluru) — hyperspectral imaging startup
- Skylo — geo-IoT (limited vision)
- Cropin — agritech satellite analysis
Security / Surveillance:
- Staqu Technologies — Indian CV company for police/security
- Vehant Technologies — vehicle surveillance
- Indian government Smart Cities Mission
Global product companies with Indian CV teams:
- Microsoft IDC — Computer vision in Azure AI
- Google India — Google Photos, Lens
- Amazon India — Rekognition, retail vision
- Apple India — privacy-preserving vision
- Adobe India — creative AI, content credentials
Indian computer vision engineer salary expectations:
- Fresher: ₹8-18 LPA at startups, ₹12-25 LPA at product companies
- 2-3 years: ₹18-40 LPA
- 5-7 years: ₹40-80 LPA
- Senior (8+ years): ₹80 LPA-1.5 Cr at top product companies
- Research scientist (PhD): ₹40 LPA-2 Cr+
Why CV pays well in India:
- Real business problems with measurable ROI
- Smaller talent pool than general ML
- Hardware integration knowledge (cameras, sensors) adds complexity
- Specific to high-value verticals (healthcare, finance, automotive)
CV is a strong specialization — pays better than general ML on average and has less competition.
Core CV skills Indian companies test for
Computer vision interviews test 4 dimensions:
1. Image processing fundamentals (still relevant):
- Color spaces (RGB, HSV, LAB, YUV)
- Filters (Gaussian, median, bilateral)
- Edges (Canny, Sobel, Laplacian)
- Morphological operations (erosion, dilation, opening, closing)
- Histograms and histogram equalization
- Affine transformations, perspective transforms
- Feature detection (SIFT, ORB, AKAZE) — increasingly replaced by deep learning but still asked
2. Classical computer vision (asked at older companies):
- Optical flow
- Stereo vision and depth estimation
- Camera calibration (intrinsics, extrinsics)
- Bundle adjustment (basics)
- SLAM (Simultaneous Localization And Mapping)
- Image stitching
3. Deep learning for vision (the primary skill in 2026):
- CNNs from scratch (understand convolution, pooling, padding)
- Modern architectures: ResNet, EfficientNet, ConvNeXt
- Object detection: YOLO family (YOLOv8, YOLOv9, YOLO11), DETR
- Semantic segmentation: U-Net, DeepLab, SAM (Segment Anything)
- Instance segmentation: Mask R-CNN, YOLACT
- Pose estimation: OpenPose, MediaPipe
- Vision Transformers (ViT) — increasingly important in 2026
- Multimodal models: CLIP, BLIP-2, LLaVA, Gemini Vision
4. Production CV (the senior gap):
- Model optimization: quantization, pruning, distillation
- Deployment: ONNX, TensorRT, OpenVINO, CoreML, TFLite
- Edge deployment: Jetson Nano/Orin, Coral TPU, ESP32-CAM
- Camera integration (RTSP, ONVIF, USB cameras)
- Real-time constraints (latency budgets)
- Multi-stream video processing
Tools every CV engineer must know:
- OpenCV (Python)
- PyTorch (recommended over TensorFlow in 2026)
- Albumentations (image augmentation)
- Ultralytics / YOLO library
- Hugging Face transformers (for ViT, CLIP, etc.)
- Roboflow for dataset management
- Weights & Biases for experiment tracking (free tier)
Indian interview question patterns:
- "Explain how YOLO works" — common at vision-heavy startups
- "How would you detect counterfeit products in our marketplace?" — Flipkart/Myntra
- "Design a face recognition system that's robust to lighting changes" — Hyperverge
- "How would you handle adversarial images?" — security companies
- "Walk through the deployment of a vision model to a mobile device" — Microsoft IDC, Adobe India
9-month computer vision roadmap (free)
Prerequisites: Python proficiency, ML basics (use our AI/ML Engineering Guide).
Months 1-2 — Image processing + OpenCV
- OpenCV Python tutorials
- Read, display, save images and videos
- Color space conversion
- Filtering, edges, contours
- Feature detection (ORB, SIFT)
- Basic projects: barcode scanner, license plate reader, document scanner
Free resources:
- PyImageSearch by Adrian Rosebrock — best free CV tutorials globally
- OpenCV official tutorials
- LearnOpenCV by Satya Mallick — Indian founder, excellent content
- Murtaza's Workshop YouTube
Month 3 — CNN fundamentals
- Convolution, pooling, padding from first principles
- Build a CNN from scratch in PyTorch for MNIST then CIFAR-10
- Transfer learning with pre-trained ResNet
- Common architectures (AlexNet → VGG → ResNet → EfficientNet)
- Image augmentation strategies (random crops, flips, color jitter, mixup, cutmix)
Free resources:
- fast.ai Practical Deep Learning — vision-first course
- PyTorch official vision tutorials
- Andrej Karpathy's CS231n Stanford course (free on YouTube) — best CV course content available
Month 4 — Object detection
- Bounding box concept, IoU (Intersection over Union)
- Non-Maximum Suppression (NMS)
- Anchor-based detectors (Faster R-CNN, YOLO v3-v5)
- Anchor-free detectors (CenterNet, FCOS, YOLOX)
- Modern: YOLOv8, YOLO11 with Ultralytics
- Build a custom object detector for an Indian-specific dataset (e.g., Indian traffic signs, food items)
Month 5 — Segmentation
- Semantic segmentation: U-Net, DeepLabV3+
- Instance segmentation: Mask R-CNN
- Foundation models: SAM (Segment Anything Model)
- Practical project: segment medical images, satellite images, or document layouts
Month 6 — Modern transformers for vision
- Vision Transformer (ViT) architecture
- DETR (DEtection TRansformer)
- Multimodal: CLIP, BLIP-2, LLaVA
- Hugging Face transformers library for vision
- Practical: build an image search engine using CLIP embeddings
Month 7 — Production CV
- Convert PyTorch to ONNX
- TensorRT for NVIDIA GPU optimization
- OpenVINO for Intel hardware
- TFLite / CoreML for mobile
- Latency benchmarking and profiling
- Quantization (INT8) for speed
Month 8 — End-to-end project
- Pick one Indian-specific problem:
- Indian crop disease detection (vary by state)
- Indian document OCR (Aadhaar, PAN, driving license)
- Indian retail shelf monitoring
- Indian traffic violation detection
- Indian medical imaging analysis
- Build dataset (or use existing — Kaggle Indian datasets, AI4Bharat)
- Train, evaluate, deploy as a web demo (Gradio or Streamlit)
Month 9 — Portfolio + interviews
- Write 2-3 blog posts on what you built
- Polish GitHub (3-4 strong CV projects)
- Apply to Indian CV companies (Hyperverge, Niramai, Qure.ai, Flipkart, Myntra)
- Practice CV interview questions
By end of 9 months you'll have skills competitive for fresher CV engineer roles at Indian product companies and startups.
Edge computer vision — high-paying Indian specialization
Running CV models on edge devices (cameras, phones, IoT) is a specialized skill that pays well in India:
Why edge CV matters in India:
- Many Indian applications can't rely on cloud (rural connectivity, latency, cost)
- Indian smart city projects require edge processing
- Manufacturing / agriculture / retail need on-device inference
- Privacy regulations favor edge processing for sensitive data (medical, surveillance)
Edge CV hardware (Indian context):
- NVIDIA Jetson Nano / Orin Nano (₹15,000-50,000) — popular for prototyping
- Google Coral (₹10,000-20,000) — TPU acceleration
- Raspberry Pi 4 / 5 (₹5,000-12,000) — entry-level edge
- ESP32-CAM (₹500-1,500) — ultra-low-cost
- Qualcomm RB5 Robotics Platform — high-end edge
- Smartphone cameras (using TFLite / CoreML)
Edge CV skills:
- Model compression: quantization (INT8, INT4), pruning, distillation
- Knowledge distillation (teacher-student)
- TFLite for Android
- CoreML for iOS
- ONNX Runtime for cross-platform
- TensorRT for NVIDIA Jetson
- OpenVINO for Intel hardware
- Custom CUDA kernels for advanced optimization
Indian companies hiring edge CV engineers:
- Qualcomm India (Bengaluru) — Snapdragon AI engines
- NVIDIA India (Pune, Bengaluru) — Jetson team
- Samsung India R&D — on-device camera AI
- Bosch India — automotive vision (must be edge for latency)
- Indian smart city projects (Surat, Indore, Bhubaneswar)
- Manufacturing automation startups (Detect Technologies, Vimaan Robotics)
Salary premium for edge CV:
- 30-50% higher than vanilla CV roles at the same level
- Fresher edge CV: ₹15-25 LPA at top companies
- Mid-level: ₹30-55 LPA
- Senior: ₹55 LPA-1.2 Cr
Free resources for edge CV:
- NVIDIA Jetson Developer Resources
- TensorFlow Lite documentation
- Google Coral docs
- PyImageSearch edge CV tutorials
Edge CV is genuinely under-supplied in India — fewer than 1000 engineers have deep edge CV expertise. If you can build credibility here, the career ceiling is high.
Research scientist path for CV in India
If you're interested in pushing the state-of-art rather than applying existing models:
Indian CV research labs:
- Microsoft Research India (MSR India, Bengaluru) — CV is one focus area
- Google Research India (Bengaluru) — limited but growing
- IBM Research India (Bengaluru, Delhi) — vision for healthcare, manufacturing
- Indian Statistical Institute (ISI Kolkata, Bengaluru) — strong vision group
- IISc (Bengaluru) — multiple vision research groups
- IIT Bombay, IIT Delhi, IIT Madras — strong vision research
- IIIT-Hyderabad — Center for Visual Information Technology (CVIT) — major Indian CV research center
- TCS Research — applied CV research
- Adobe Research India — generative vision
Typical path to research scientist:
- B.Tech CS / EE / Math with strong academics
- M.Tech / MS in CS at top Indian institute (IIT, IIIT-H, IISc) OR international (CMU, Stanford, etc.)
- PhD at Indian or international university (4-6 years)
- Research scientist role: ₹40-80 LPA fresher at MSR India, Google India
Alternative paths to research-adjacent roles:
- B.Tech + 4-5 years strong industry research (publications + open source contributions) → can compete for research roles
- M.Tech with research focus (CVIT, IIT vision groups) → applied research roles
- Self-taught with strong portfolio + paper contributions → niche research roles
Required for research career:
- Strong math background (probability, optimization, linear algebra)
- Ability to read 50+ papers/year and synthesize
- Code reproduction skills (can you reimplement a paper from its description?)
- Writing skills (papers, internal docs)
- Conference publication track (NeurIPS, CVPR, ICCV, ECCV, ICLR — top tier)
Indian CV research output:
- IIT/IIIT/IISc collectively publish 50-100+ papers/year at top vision conferences
- Indian researchers globally publish 5-10% of CVPR/ICCV papers
- Strong tradition in low-resource vision (training with limited data), Indian-language OCR, healthcare AI
Free resources to follow:
- arxiv.org/cs.CV — daily computer vision papers
- Two Minute Papers YouTube — paper summaries
- Yannic Kilcher YouTube — deep paper walkthroughs
- Papers with Code — papers + benchmarks + code
Research scientist is a smaller but lucrative path. Most Indian engineers will be CV engineers (applied) rather than research scientists. Both are valid.
Frequently asked questions
Can I become a CV engineer without an ML background?▾
You'll need ML fundamentals (3-6 months prerequisite). CV is a specialization of ML — without understanding general ML (gradient descent, overfitting, evaluation), CV will feel like magic. Plan: 3 months ML basics, then 6-9 months CV specialization.
OpenCV or PyTorch for first CV library?▾
Both — learn together. OpenCV for image processing fundamentals (filters, transforms, classical CV). PyTorch for deep learning models. Most Indian CV engineers use both daily — OpenCV for preprocessing/postprocessing, PyTorch for the deep learning model.
Is YOLO still relevant in 2026?▾
Absolutely yes. YOLOv8, YOLO11, YOLO-NAS are state-of-art for real-time object detection. YOLO family remains the most-deployed object detector globally including in India. Learn it deeply — both how it works internally and how to fine-tune it on custom data.
Computer vision or NLP — which has more Indian jobs?▾
NLP slightly more in 2026 due to LLM boom, but CV is more stable. CV jobs are spread across more industries (healthcare, automotive, retail, security, manufacturing). NLP is concentrated in tech (LLM startups, search, recommendations). Both pay similarly at senior levels.
What hardware do I need to learn CV?▾
A laptop with 16GB RAM can do most learning. For training larger models, use Google Colab free tier (T4 GPU, 12-hour sessions) or Kaggle notebooks (P100 GPU). For deployment learning, an NVIDIA Jetson Nano (₹15,000) is the cheapest GPU device. You DON'T need an expensive PC to start.
What's the best free CV course in 2026?▾
Stanford CS231n on YouTube (Karpathy's lectures specifically) for theory. PyImageSearch for hands-on tutorials. fast.ai for project-first learning. Combine all three over 6 months. Free Indian resources: LearnOpenCV by Satya Mallick is exceptional.
How important is mathematics for CV?▾
Moderately important. You need: linear algebra (matrices, vectors, transformations), calculus (gradients, backprop), probability basics. Don't need: measure theory, functional analysis. The math bar for CV engineer is similar to general ML engineer — solid but not PhD-level.