COMPUTER VISION
ImageIQ — Computer Vision Analytics Platform
Dataset annotation tool: upload images, pre-label them with YOLOv8-n detection, search the library by text with CLIP ViT-B/32 embeddings, review predictions in a queue, and export annotations as COCO JSON, YOLO TXT, Pascal VOC XML or CSV.
Public demo
4 — annotation export formats
The problem
Exploring and labelling an image dataset usually means scrolling folders and drawing every box by hand. There is no quick way to ask for 'images of X' or to pre-label with a detector and review the result instead.
The approach
A FastAPI backend runs YOLOv8-n (the ~6 MB nano model, CPU-only) for detection and CLIP ViT-B/32 for text-to-image search over a FAISS index, behind a multi-step batch pipeline (preprocess → inference → embed → annotate) with polled progress. The heavy model dependencies are optional: without them the API falls back to simulated detections so every other route and the full UI still work, which keeps the free-tier deployment within its memory limit.
The results
Auto-labelling with a human review queue (approve / edit / reject) and four export formats. Detection quality is that of the pretrained YOLOv8-n — no dataset-specific training or benchmark was run, and the public demo may be using the simulated fallback.
Stack: React 18, TypeScript, FastAPI, YOLOv8-n, CLIP ViT-B/32, FAISS, Python 3.11