~/hailcpy
all projects
researchmlcomputer-vision

Comparing DNN with Human Representations

Behavioural experiment comparing CNN-based similarity predictions to human similarity judgments. Won Best Research Award at SnT Council, IIT Kanpur.

What it is

Research project asking: when a CNN says two images are similar, does that match what humans say? Designed a behavioural experiment and compared CNN feature similarity to human judgments across animal images.

What I shipped

CNN Feature Extraction

Extracted deep image features using pre-trained network backbones (VGG16, ResNet) to compute similarity matrices.

Behavioural Platform

Designed and shipped a custom web utility in HTML/CSS/JS to collect human similarity judgments on the same image pairs.

Statistical Comparison

Conducted alignment analyses between neural network feature vectors and human similarity judgements.

Project Info

TimelineApr 2020 — Jul 2020

Key Metrics

award
Best Research · SnT
venue
BCS · IIT Kanpur

Technology Stack

researchmlcomputer-vision