ummanmm/classroom-reaction-vgg16

Classroom Reaction Recognition - VGG16

A VGG16 classifier trained to recognize facial reactions of students from cropped person bounding boxes in classroom lecture videos.

Classes

Index	Label	Description
0	Neutral	No visible expression
1	Confused	Furrowed brow, squinting
2	Smiling_Amused	Visible smile, laughter
3	Surprised	Raised eyebrows, open mouth
4	Bored_Tired	Yawning, blank stare

Architecture

• Backbone: VGG16 (ImageNet pre-trained, feature layers frozen)
• Classifier head: nn.Linear(4096, 5)
• Loss: Weighted CrossEntropyLoss to handle class imbalance
• Optimizer: Adam (lr=0.001)
• Input size: 224 x 224 RGB, ImageNet-normalized

Training

Trained on manually annotated person crops extracted from classroom lecture videos using YOLOv8-nano detection. 70/10/20 stratified train/val/test split.

Usage

import torch
from torchvision import models

model = models.vgg16(weights=None)
model.classifier[6] = torch.nn.Linear(4096, 5)
model.load_state_dict(torch.load("best_vgg16.pth", map_location="cpu"))
model.eval()