Course Objective & Scope
For area measurement, defect boundaries, tissue analysis or land classification, a bounding box is not enough; you need the class of every pixel. This programme focuses solely on segmentation and covers semantic, instance and panoptic tasks in depth through U-Net, DeepLabV3+, SegFormer and Mask2Former.
Prompt-based segmentation and faster labelling with SAM 2, choosing Dice and Focal loss, boundary quality, tiling for high-resolution images and area/length measurement from masks are practised in hands-on labs. Participants design a segmentation pipeline suited to their own data types.
Who Should Attend?
- Computer vision engineers on projects requiring pixel-level analysis
- Industrial quality, medical imaging or remote sensing teams
- Data scientists moving from detection models to segmentation
- Prerequisite: experience training CNN models with PyTorch
Key Takeaways
- Distinguish semantic, instance and panoptic tasks per problem
- Train a custom model with U-Net, SegFormer or Mask2Former
- Shorten labelling time with SAM 2
- Evaluate models with Dice, mIoU and boundary metrics
- Produce area and size measurements from masks
Curriculum
Segmentation Tasks and Use Cases
- Semantic, instance and panoptic differences
- Industrial, medical and satellite imagery examples
Encoder-Decoder Architectures
- U-Net and DeepLabV3+ structures
- Skip connections and atrous convolution
Transformer-Based Segmentation
- SegFormer and Mask2Former
- Fine-tuning with Hugging Face Transformers
Prompt-Based Segmentation with SAM 2
- Generating masks from point and box prompts
- Faster labelling and video segmentation
Loss, Metrics and Post-Processing
- Dice, Focal loss and class imbalance
- mIoU, boundary F1 and measurement from masks
High Resolution and Production
- Tiling, sliding windows and memory management
- Inference speed and ONNX export
Other Courses in This Category
-
ViT, DeiT, Swin and DINOv2 architectures: patch embedding, self-attention, comparison with CNNs and fine-tuning with Hugging Face, for engineers selecting and adapting transformer-based models for vision tasks.
- Live Course
- 5 modules
-
Document preprocessing, text recognition, layout analysis and table/form extraction with PaddleOCR, Surya, LayoutLMv3 and Donut, for technical teams turning invoices, contracts and operational documents into structured data.
- Live Course
- 5 modules
-
RTSP video streams, GStreamer/FFmpeg, real-time detection, multi-object tracking with ByteTrack/BoT-SORT, speed/crowd analytics and NVIDIA DeepStream, for teams extracting live operational insights from camera networks.
- Live Course
- 5 modules
-
Human action, ergonomics and anomaly detection with 3D CNNs (SlowFast, X3D), VideoMAE/TimeSformer and skeleton-based models (MMPose, ST-GCN), for teams building workplace safety, assembly-line and behaviour analysis systems.
- Live Course
- 5 modules
-
Shared embedding spaces with CLIP/SigLIP, Qwen2.5-VL, LLaVA and InternVL architectures, visual QA, multimodal RAG and QLoRA fine-tuning, for engineers building systems that reason across images and text.
- Live Course
- 5 modules
-
Surface defect detection on production lines, unsupervised anomaly detection with anomalib (PatchCore, PaDiM, EfficientAd), industrial camera/lighting design and PLC integration, for quality and manufacturing engineering teams.
- Live Course
- 5 modules
Get in touch
