Searchable datasets in Python
summary
A Python package called pyCBIR performs content‑based image retrieval across domains, exporting signatures to CSV and offering GUI and CLI. It implements six feature sets (GLCM, HOG, LBP, texture, histogram, CNN), ten distance metrics, and two CNN workflows: train‑then‑extract (example: training ~3 min on 16×16 fibrous set, feature extraction 9 s, 4 s to retrieve top‑10) versus pretrained inception (extraction 29 min, 50 s retrieval). Experiments on ~3–4k images and a 3.5k cervical set show workable results on a laptop with clear tradeoffs in time, signature size and accuracy.
AI-written summary · about
chapters (21)
- 0:00Searchable image datasets
- 2:12PicSciBeer and core pillars
- 4:02Content-based image retrieval history
- 5:57Open-source vision for PicSciBeer
- 7:22Need for PicSciBeer in imaging
- 8:49Signatures and similarity metrics
- 11:20Image signatures and feature vectors
- 12:07Feature extraction methods
- 14:23Histogram of oriented gradients
- 15:48Texture features and metrics
- 16:34Local binary patterns
- 17:16Convolutional neural networks in PicSciBeer
- 19:04Comparing signatures with metrics
- 20:51PicSciBeer GUI setup
- 22:40Command-line demo walkthrough
- 24:31Experiment with fibrous tissue
- 26:40Experiment with cervical cells
- 28:30Training approaches comparison
- 29:35Next steps and improvements
- 31:25Future challenges
- 33:30Closing and hiring
related talks
37mData Science for Digital Humanities: Extracting meaning from Images and Text
01h 29mHuman-like Visual Search Application with Small Data
22mOCR, Recognition and Information Extraction from Images
18mCustomizable Computer Vision Expands Data Access Without Compromising Privacy
40m"Introduction to Face Processing with Computer Vision|"
28mWhy I didn't use deep learning for my image recognition
30mThe Art and Science of Deep Face Recognition in Python
31mOpenDataCube - Petabyte Scale Satellite Imagery Processing in Python