Weilin Xu

It would be nice to flatten so we get per class AP numbers. This doesn't work right now though and just returns the average 🙃 _Originally posted by @dxoigmn in...

Inconsistency between configuration and code in the Missed objective of object detection

1

https://github.com/IntelLabs/MART/blob/6184ca764c871cc098a4e822e520f476e556a971/mart/configs/attack/objective/object_detection_missed.yaml#L5-L7 https://github.com/IntelLabs/MART/blob/6184ca764c871cc098a4e822e520f476e556a971/mart/attack/objective/object_detection.py#L68 The code requires `preds` and `target`, but the configuration only gives `preds`. Since `target` is not used in the objective function, we can either delete it from `__call__()`'s...

# What does this PR do? This PR adds support to object detection adversary. As the implementation is modality-aware, it should not be hard to implement RGB-Depth adversary. ```bash mart...

# What does this PR do? This PR makes external attackers part of the unified `Adversary` object which enforces threat models. ## Type of change Please check all relevant options....

Python's built-in function `input` is overriden by a variable

1

For example, https://github.com/IntelLabs/MART/blob/2c62aad375e146036696e88c09f5cb3a0f7131fd/mart/models/modular.py#L103 This is not good, because static analysis tools may not understand the semantics of `input` in the code.

We need to make it easier to wrap models to be evaluated in ART.

# What does this PR do? This PR adds an example of constructing adversaries in Anomalib. See [examples/anoamlib_adversary/README.md](https://github.com/IntelLabs/MART/tree/mzweilin/example_anomalib_adversary/examples/anomalib_adversary) for details. ## Type of change Please check all relevant options. -...

## 📝 Description - This PR visualizes images that are already loaded in a batch, instead of loading them again from the file system. This change allows visualizing adversarial images...