Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix UFlow links #2059

Merged
merged 2 commits into from
May 16, 2024
Merged

Fix UFlow links #2059

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
135dbb4
fix links
ashwinvaidya17 May 15, 2024
8c6e3d3
Merge branch 'main' into fix/uflow_links
ashwinvaidya17 May 16, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions src/anomalib/models/image/uflow/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,15 +40,15 @@ Nevertheless, the code for computing the segmentation mask with the NFA criterio

## Benchmarking

Note that the proposed method uses the MCait Feature Extractor, which has an input size of 448x448. In the benchmarking, a size of 256x256 is used for all methods, and therefore the results may differ from those reported. In order to exactly reproduce all results, the reader can refer to the original code (see [here](https://github.com/mtailanian/uflow), where the configs used and even the trained checkpoints can be downloaded from [this release](https://github.com/mtailanian/uflow/releases/tag/trained-mvtec-models).
Note that the proposed method uses the MCait Feature Extractor, which has an input size of 448x448. In the benchmarking, a size of 256x256 is used for all methods, and therefore the results may differ from those reported. In order to exactly reproduce all results, the reader can refer to the original code (see [here](https://github.com/mtailanian/uflow), where the configs used and even the trained checkpoints can be downloaded from [this release](https://github.com/mtailanian/uflow/releases/tag/trained-models-for-all-mvtec-categories).

## Reproducing paper's results

Using the default parameters of the config file (`src/anomalib/models/uflow/config.yaml`), the results obtained are very close to the ones reported in the paper:

bottle: 97.98, cable: 98.17, capsule: 98.95, carpet: 99.45, grid: 98.19, hazelnut: 99.01, leather: 99.41, metal_nut: 98.19, pill: 99.15, screw: 99.25, tile: 96.93, toothbrush: 98.97, transistor: 96.70, wood: 96.87, zipper: 97.92

In order to obtain the same exact results, although the architecture parameters stays always the same, the following values for the learning rate and batch size should be used (please refer to the [original code](https://github.com/mtailanian/uflow) for more details, where the used configs are available in the source code ([here](https://github.com/mtailanian/uflow/tree/main/configs)), and trained checkpoints are available in [this release](https://github.com/mtailanian/uflow/releases/tag/trained-mvtec-models)):
In order to obtain the same exact results, although the architecture parameters stays always the same, the following values for the learning rate and batch size should be used (please refer to the [original code](https://github.com/mtailanian/uflow) for more details, where the used configs are available in the source code ([here](https://github.com/mtailanian/uflow/tree/main/configs)), and trained checkpoints are available in [this release](https://github.com/mtailanian/uflow/releases/tag/trained-models-for-all-mvtec-categories)):

## Usage

Expand Down
Loading