LINC: AI for the conservation of the African lion
summary
LINC builds a working pipeline: a Faster R-CNN object detector for faces and whisker regions, a ResNet-50 trained classifier turned 512-d embedding with k‑NN retrieval, and a whisker-point approach using coherent point drift for pattern matching. Dataset: ~400 lions, ~6,000 images (filtered to ≥10 images per individual, ~350+ left); initial top‑5 accuracy leaked to 95% due to near-duplicate splits, corrected evaluation on 20 held-out lions yields ~85% top‑5 (top‑10/20 >95%). Object-detection label/augmentation bugs fixed; production-ready UI and database integrated.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:08Agenda
- 3:35Manual lion identification
- 4:17Their proposal
- 4:59Timeline
- 6:21What did the data look like?
- 8:07Faster R-CNN
- 8:34Initial results were looking great!
- 10:36Finally, consistent results!
- 11:03We even tested on CGI lions
- 11:29Known issues
- 11:56Using the same ideas on whiskers
- 13:47Identification data
- 14:33Image preprocessing
- 16:33Building the database
- 17:54Getting a prediction
- 21:35Coherent point drift on whiskers
- 22:10Where it fails
- 23:06Lines of work
- 23:27Combining both approaches
related talks
12mWill AI help save the snow leopard?
24mImage Recognition for safety on the factory floor
25mComputer Vision in Melanoma Detection
33mBuilding end-to-end computer vision solutions from deep learning models
01h 29mHuman-like Visual Search Application with Small Data
01h 15mTrain Object Detection with Small Datasets
32mYou Can Do Deep Learning!
52mFace Recognition: From Scratch To Hatch