24679-hw2-stairs-autogluon

ypolatog/24679-hw2-stairs-autogluon

24679-hw2-stairs-autogluon is a image classification model from ypolatog. Use it when you need a label for an image. It is set up for timm. The card lists the license as other.

Binary image classifier built with AutoGluon MultiModalPredictor over timm backbones, trained on ArinRoths/StairvsNonStairDataset, a classmate's HW1 dataset.

Downloads · 30 days

Access

Public

Updated Sep 21, 2026

Repo size

165 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.pkl85.4 MB · 52%

At a glance

Task: Image Classification
Library: timm
License: other
Access: Public
Created: Sep 21, 2026
Updated: Sep 21, 2026
SHA: 578690d5

Model card

From the Hugging Face model README

Stairs vs non-stairs classifier — 24-679 HW2 Part 2

Binary image classifier built with AutoGluon MultiModalPredictor over timm backbones, trained on ArinRoths/StairvsNonStair_Dataset, a classmate's HW1 dataset.

Results

Metric	Cross-validated (3-fold grouped)	Held-out test (n=5)
Accuracy	0.444 +/- 0.050	0.600
Balanced accuracy	0.456	0.583
F1 (stairs)	—	0.500

Chance is 0.500. Majority baseline on test: 0.400 (always predicts stairs).

Selected configuration

Chosen from 10 random-search trials by validation ROC AUC (0.938), each capped at 180 s with early stopping after 3 checks without improvement.

Hyperparameter	Value
`backbone`	`resnet34`
`image_size`	`224`
`augmentation`	`none`
`optimizer`	`sgd`
`learning_rate`	`7.348782489930548e-05`
`weight_decay`	`0.0002499845149337477`
`lr_decay`	`0.6`
`batch_size`	`32`

Architectures searched: resnet18, resnet34, efficientnet_b0, mobilenetv3_large_100.

Intended use and limitations

Coursework for CMU 24-679 Design AI; not for any real decision. Trained on 27 original photographs taken by one person, so it reflects that camera, those locations and that lighting. The held-out test split is 5 photographs — far too few to judge on alone, which is why the cross-validated figure is quoted first.

The source dataset declares no license. Reuse should be cleared with its author, ArinRoths.

Reproducing

Seed 24679, AutoGluon multimodal 1.6.1. metrics.json holds every number above and search_results.csv holds all 10 trials. Unzip autogluon_image_predictor_dir.zip and load with MultiModalPredictor.load().

Task: Image Classification
Library: timm
License: other
Created: Sep 21, 2026
Updated: Sep 21, 2026