Skip to content

Evaluation of Inception Score and default config of DDT-B/2 #11

@Nudavia

Description

@Nudavia

I use the default config of DDT-XL/256 to sample images and download VIRTUAL_imagenet256_labeled.npz for evaluation using 'ADM-evaluation-suite-pytorch'
This is my script:

pred_npz=universal_flow_workdirs/exp_repa_flatten_condit22_dit6_fixt_xl_20251110_014343/val/predict/output.npz

python ADM-evaluation-suite-pytorch/evaluator.py \
    /node2/disk2/dataset/ILSVRC2012/VIRTUAL_imagenet256_labeled.npz \
    ${pred_npz}

The results:

Inception Score: 67.99991607666016
FID: 1.2773754512323308
sFID: 4.532291743289761
Precision: 0.78816
Recall: 0.6442

Most are right except Inception Score, why is the score so low ?
By the way, can you show the config of DDT-B/2 please?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions