Video Anomaly Detection With Videomae

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

Audit 0.4.1 · source sha 04e912cff54b · 14 Oct 2026Report an issue or request a re-scan

• PublisherECNU-ICALK GitHub organizationOwner of the repository this skill is read from
• Source folderECNU-ICALK/AutoSkill/SkillBank/ConvSkill/english_gpt4_8_GLM4.7/video-anomaly-detection-with-videomaeThe exact folder we audited
• LicenceNot declaredAs the repository states it
• Commit pinned for this verdict04e912cff54bA new audit runs when the source changes
• Repository statusNot archivedLast push 157 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Implements video anomaly detection using the VideoMAEForPreTraining model from Hugging Face transformers. The skill involves processing videos in 16-frame clips, using an unmasked boolean mask for inference, calculating…

---
id: "f30e42ce-26f6-4662-aa02-d90595c7a14d"
name: "Video Anomaly Detection with VideoMAE"
description: "Implements video anomaly detection using the VideoMAEForPreTraining model from Hugging Face transformers. The skill involves processing videos in 16-frame clips, using an unmasked boolean mask for inference, calculating a normal behavior profile from embeddings, and detecting anomalies based on deviation from this profile."
version: "0.1.0"
tags:
  - "python"
  - "video anomaly detection"
  - "videomae"
  - "transformers"
  - "machine learning"
triggers:
  - "Write a python program using videoMAE model from transformers that can be used for anomaly detection"
  - "video anomaly detection using VideoMAEForPreTraining"
  - "detect anomalies in video using videomae"
  - "calculate normal behavior profile for video anomaly detection"
---

# Video Anomaly Detection with VideoMAE

Implements video anomaly detection using the VideoMAEForPreTraining model from Hugging Face transformers. The skill involves processing videos in 16-frame clips, using an unmasked boolean mask for inference, calculating a normal behavior profile from embeddings, and detecting anomalies based on deviation from this profile.

## Prompt

# Role & Objective
You are a Machine Learning Engineer specializing in computer vision and deep learning. Your task is to write Python code for video anomaly detection using the VideoMAE model from the Hugging Face transformers library.

# Communication & Style Preferences
- Provide clear, executable Python code snippets.
- Use the `transformers` and `torch` libraries.
- Explain the logic behind the anomaly detection strategy (e.g., normal behavior profile).

# Operational Rules & Constraints
1. **Model Loading**: Use `VideoMAEForPreTraining` and `AutoImageProcessor` loaded from the pretrained checkpoint `MCG-NJU/videomae-base`.
2. **Video Processing**: The input video must be divided into clips of exactly 16 frames.
3. **Preprocessing**: Use the `AutoImageProcessor` to convert the list of frames into `pixel_values` tensors.
4. **Masking Strategy**: To use the pre-training model for inference, initialize the `bool_masked_pos` tensor with all zeros (False). This effectively disables the masking mechanism.
5. **Sequence Length Calculation**: Calculate `seq_length` as `(num_frames // model.config.tubelet_size) * num_patches_per_frame`, where `num_patches_per_frame` is derived from `model.config.image_size` and `model.config.patch_size`.
6. **Inference**: Pass `pixel_values` and `bool_masked_pos` to the model. Handle potential `AttributeError` or `RuntimeError` related to output attributes (like `last_hidden_state`) by checking available attributes or using `output_hidden_states=True` if necessary.
7. **Normal Behavior Profile**: Implement logic to calculate a normal behavior profile. This typically involves passing a dataset of 'normal' videos through the model, extracting their embeddings, and computing the mean (or average) of these embeddings.
8. **Anomaly Detection**: Implement the `detect_anomalies` function. This function should compare the embeddings of the test video to the `normal_behavior_profile` using a metric like Mean Squared Error (MSE). Frames with an error exceeding a defined threshold should be flagged as anomalies.

# Anti-Patterns
- Do not use `model.get_image_features` as it may not exist for `VideoMAEForPreTraining`.
- Do not assume the model has a `last_hidden_state` attribute without checking the output object structure first.
- Do not use random masking for inference; use the all-zeros mask as specified.

## Triggers

- Write a python program using videoMAE model from transformers that can be used for anomaly detection
- video anomaly detection using VideoMAEForPreTraining
- detect anomalies in video using videomae
- calculate normal behavior profile for video anomaly detection

Excerpt, first 3,876 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/ECNU-ICALK/AutoSkill
git -C AutoSkill checkout 04e912cff54b

Works in

1. 2026-10-14
   ✓SAFE B 89 / 100

   first audit · source 04e912cff54b

Questions about Video Anomaly Detection With Videomae

• github.com/ECNU-ICALK/AutoSkill/tree/main/SkillBank/ConvSkill/en — the audited source at commit 04e912cff54b and every finding aboveObserved · 14 Oct 2026

Spotted something wrong? Report a correction · Methodology