Skip to content

Latest commit

 

History

History
96 lines (83 loc) · 3.98 KB

dataset_instructions.md

File metadata and controls

96 lines (83 loc) · 3.98 KB

CCTV-GUN setup

In this file, we provide instructions on how to download and construct CCTV-GUN benchmark.

Requirements

Some images in the original MGD source have byte code errors. Our script to fix them requires a couple of python packages. To install them, run

pip3 install -r requirements/dataset.txt

Downloading images

We ask the user to download the images of MGD, USRT and UCF from original source first.

MGD

  • Download MGD.rar file from the original source link and extract into the current directory.
    • The extracted folder would have the following structure :
    MGD
├── annotations_cache
├── MGD2020
│   ├── Annotations
│   ├── ImageSets
│   │   └── Main
│   └── JPEGImages
└── results
    ├── VOC2007
    │   └── Main
    └── VOC2020
        └── Main
  • Run python3 scripts/copy_images_mgd.py to copy images used in CCTV-GUN

USRT

  • Download weapons_images_2fps.zip file from the original source link.
    • Run unzip weapons_images_2fps.zip
    • A folder called Images will be extracted into the current directory.
  • Run python3 scripts/copy_images_usrt.py to copy images used in CCTV-GUN

UCF

  • Download Anomaly-Videos-Part-3.zip from the original source link.
    • Run unzip Anomaly-Videos-Part-3.zip
    • A folder called Anomaly-Videos-Part-3 will be extracted into the current directory.
  • Run python3 scripts/copy_images_ucf.py to copy images used in CCTV-GUN. Since UCF dataset has videos, we extract specific frames and save them as JPG images. The frame ids can be found in data/ucf/frames.json

After downloading and extracting all three datasets, the data folder should look like this

data
├── all_images #Contains images from all 3 sources
├── dataset_pairs
│   ├── mgd_usrt
│   │   ├── annotations_all.json
│   │   ├── annotations_test.json
│   │   ├── annotations_train.json
│   │   └── annotations_val.json
│   ├── ucf_mgd
│   │   ├── annotations_all.json
│   │   ├── annotations_test.json
│   │   ├── annotations_train.json
│   │   └── annotations_val.json
│   └── usrt_ucf
│       ├── annotations_all.json
│       ├── annotations_test.json
│       ├── annotations_train.json
│       └── annotations_val.json
├── mgd
│   ├── annotation_detection
│   │   ├── annotations_all.json
│   │   ├── annotations_test.json
│   │   ├── annotations_train.json
│   │   └── annotations_val.json
│   └── images 
├── ucf
│   ├── annotation_detection
│   │   ├── annotations_all.json
│   │   ├── annotations_test.json
│   │   ├── annotations_train.json
│   │   └── annotations_val.json
│   ├── frames.json
│   └── images
└── usrt
    ├── annotation_detection
    │   ├── annotations_all.json
    │   ├── annotations_test.json
    │   ├── annotations_train.json
    │   └── annotations_val.json
    └── images

Annotations

We provide bounding box annotations of all images in CCTV-GUN dataset in data/<split>/annotations_detection. Train-val-test split can be found in the same folder. The annotations are in MS-COCO format, and also contain the person-handgun pair.

To train models on pairs of datasets (Ex: MGD + USRT), the annotations provided in data/dataset_pairs/<split> can be used.