Validated case study · Deep Learning

Humach Vision: Image Classification Audit

Audit image metadata, split integrity and class balance before training a vision model.

Download mini_vision.csvUse Python, SQL, Excel or your preferred workflow unless a task specifies otherwise.
Project mode
Project dataset

Inspect the CSV before you answer the tasks

Scroll horizontally or vertically to examine the exact miniature data used by the validator.

Download CSV
image_idclasswidthheightbrightnesssplit
IMG001dog2242240.38train
IMG002bird2242240.46train
IMG003cat2242240.54validation
IMG004dog2242240.62test
IMG005bird3202240.7train
IMG006cat2242240.78train
IMG007dog2242240.3train
IMG008bird2242240.38validation
IMG009cat2242240.46test
IMG010dog3202240.54train
IMG011bird2242240.62train
IMG012cat2242240.7train
IMG013dog2242240.78validation
IMG014bird2242240.3test
IMG015cat3202240.38train
IMG016dog2242240.46train
IMG017bird2242240.54train
IMG018cat2242240.62validation
IMG019dog2242240.7test
IMG020bird3202240.78train
IMG021cat2242240.3train
IMG022dog2242240.38train
IMG023bird2242240.46validation
IMG024cat2242240.54test
IMG025dog3202240.62train
IMG026bird2242240.7train
IMG027cat2242240.78train
IMG028dog2242240.3validation
IMG029bird2242240.38test
IMG030cat3202240.46train
IMG031dog2242240.54train
IMG032bird2242240.62train
IMG033cat2242240.7validation
IMG034dog2242240.78test
IMG035bird3202240.3train
IMG036cat2242240.38train
IMG037dog2242240.46train
IMG038bird2242240.54validation
IMG039cat2242240.62test
IMG040dog3202240.7train
40 rows6 columnsmini_vision.csv
Project workspace

Analyse → validate → submit.

Numeric and decision tasks are auto-checked. Written reasoning is checked for completeness and key ideas, then included in your local submission report.

0/5 validated
01

How many image metadata rows are provided?

02

How many images are in the test split?

03

How many images are labelled dog?

04

What is a dangerous split issue for near-duplicate images?

05

Name one augmentation that could be reasonable and one that might change label semantics.

Submit your project locally

HumachLearn stores no server-side data. Submission creates a portable JSON report containing your answers, validation state and timestamp.