Indian Retail Vision Startups - #Infilect #Wesense #Tangoeye

 

Keep Exploring!!!

Vision - Webinar Notes - Life of Vision Project - OpenCV - Labelmg - Object Detection - Tensorflow Lite - KfServing - MySQL

Some talks summarize all of your work, plus talk on those aspects which you did implicitly part of your work. Nice Talk to connect with Vision Work :)

Key Notes

• Workflow for Image Solutions
• Dataset preparation
• Data Annotation
• Training / Benchmark
• Pre-trained + Transfer Vs Custom Model
• Metrics for benchmarking

Data Collection

• Data set type
• Streaming or Image
• Data formats
• Single image / frames
• Video - Frame - Feed model
• Image Resolution / Frame rates sampling
• Reduce frame rate to support more streams
• Preprocessing work
• Crop noisy areas
• Select areas of interest
• Data Generation
• Data Augmentation
• Simple techniques including vision tricks - rotation, transformation, different angles
• GAN / Synthetic data generation techniques

Data Annotate 

• Annotate / Review
• Validate with SME
• Bounding boxes / Segmentation / Labels
• Single / Multiple objects / Classes
• Occlusion, Light, settings
• Partially available surfaces
• Fine-grained annotation or not
• Data set representation against bias
• Coverage of possible classes
• Models for Day time vs Night time

Model Training

• Segmentation / Custom Detection
• Post-processing
• Transfer Learning

Model Optimization

• Prune / Quantize
• Inference Engines
• CPU / GPU / FPGA

Benchmark

• Testing on Deployable hardware
• Number of endpoints
• Load vs Response
• Re-annotate / Re-train
• Ensemble or Single Model

Deploy

• Edge vs Cloud 
• Edge Server - Lite weight models
• Address based on the constraint, workloads for edge devices
• Hybrid approach both edge + cloud
• Model interface with application
• Storing Results in DB
• Real-time notification or just store

Model Monitoring

• Monitoring for data/accuracy of detections
• Pick low accuracy results / retrain them
• Capture when confidence is less than 50%
• Continuous re-learning 

End to End platform for this

Keep Connecting the Dots!!!

AI City Challenge - Key Lessons

AI City Challenge - Key Lessons

2018 AI City Challenge

Key Notes

• Detection Models - YOLO2, DenseNet,Mask
• R-CNN,  Faster R-CNN
• The Mask R-CNN model, in particular, was
• able to detect and localize small vehicles with excellent precision
• Tracking - clustering-based association,  graph optimization, medianflow, Kalman filtering
• Most successful approaches are based on traffic motion flow analysis (e.g., using optical flow) rather than trying to detect and track individual vehicles
• Re-identification matching -  triplet loss
• Re-id using Vehicle number match

• Data Issues - video quality, illumination and environmental conditions

The 2019 AI City Challenge

Key Notes

• Extracting visual features from convolutional neural networks (CNNs), and leveraging semantic features from traveling direction and vehicle type classification.
• The utilization of vehicles semantic attributes 
• Novel two-stage framework based on anomaly candidate identification and starting time estimation
• Data Issues - vehicle-based problems are more challenging, due to the high intra-class variability caused by the dependence of shapes on viewing angles, and high inter-class similarity, as vehicle models produced by different manufacturers look visually alike

The 4th AI City Challenge

Key Notes

• For vehicle detection, most teams [5, 1, 37] selected YOLOv3
• For vehicle tracking, DeepSORT
• Hungarian matching algorithm to associate detections into tracklets, considering both spatial and appearance features.
• Motion-based tracking
• Trained classifiers for vehicle type, color, and viewpoint/orientation using the labels on synthetic data and made predictions on real-world data
• Multi-target single-camera (MTSC) tracking,
• ReID for appearance feature extraction, and spatio-temporal association to assign identities to tracklets across multiple cameras.
• Anomaly prediction module used K-means clustering to identify potential anomalous regions
• Data Challenges Challenges in this regard include the variety in camera views, image quality, lighting, and weather conditions

Keep Thinking!!!