1- Emotion Detection
Description
This dataset contains 6 classes of human facial emotions. Dataset was collected from employees at KICS UET Lahore and from students of Computer Science and Engineering department. Dataset covers two background scenarios that are simple and complex.
Resolution: HD
Number classes with total instances
2- Gender Dataset
Description
IBM admission dataset and google images
Number of instances
- Male – 2600
- Female – 2600
3- Sun Glasses Dataset
Description
Male and Female sun glasses dataset collected from KICS employees as well from internet.
Number of instances
- Male – 145
- Female – 24
4- Video Captioning Dataset
Description
This dataset has been collected from YouTube. Downloaded videos from YouTube were clipped into 2-5 sec segments of no transition in any attribute of video. Five frames from each video were extracted after regular intervals. Human resources were utilized to write descriptions across each frame using predefined dictionary of verbs, adjectives and objects.
Number of instances:
- Videos 2000
- Images 10,000
- Captions 10,000
5- Vehicle Crowd Heat map Generation
Description
Number of Classes: 3
Categories: Low, Medium, High
Number of Instances: 500 each class
File Ext: .npy
6- Crowd Type Dataset
Description
Number of classes: 5
Categories: Bird crowd, human crowd, animal crowd, vehicle crowd, Mix crowd
Number of Instances: 300 each class
Based on simple images not per-process image
7- Crowd Category
Description
Number of classes: 4
Categories: Concerts, uniform, sports, Violence
Number of instances: 250 each class
8- Vehicle Traffic Dataset
Description
Number of classes: 3
Categories: Low traffic , Medium Traffic, High Traffic
Number of Instances: 300 each class
Based on simple images not per-process image
9- Weather Classification Dataset
Description
Number of Classes: 3
Categories:Fog, Rain, Clear
Number of Instances: 250 each class
Based on simple images not per-process image
10- Accident Detection
Description
Categories: Accident
Number of Instances: 524
File: .XML
11- Gender Detection Dataset
Description
Number of Classes: 2
Categories: Male, Female
Number of Instances: 200 each class
File: .XML
12- Human Crowd Dataset
Description
Number of Classes :3
Categories: Individual, Row, Group
Sample :1150 Each class
13- Vehicle Detection in Parking dataset
Description
Number of Classes :1
Category: vehicle
Number of instances : 600
File: .XML
14- Occupational human Dataset Details
Dataset Details
The dataset generated for Occupational human comprises of 5 professions. Our source for collecting the data is online free sourced images and videos available on the INTERNET. We have generated image files using online news sources and saved it with unique assigned them unique name with ID.
Human resources
I scrapped and collected them from google . I saved each image in a folder with feature as follows:
- Clear images with complete human in it
- Less background noise
Occupational Categories
Naming Convention
We have three separate folders for each occupational human, with total 5 different folders each file is of jpeg file format.
Collected Number of Samples
- 780 (5 categories)
15- Face Morphing Details
Dataset Details
The dataset generated are for face morphing , comprising human face with its face key points. Our source for collecting data is online videos on INTERNET(youtube videos).
We have frames extracted form a live conference of imran khan and are saved in a folder.
Human resources
One internee scrapped and collected the Imran_Khan’s dataset from Youtube website in few days(only 4000 sample images). Each image is scrapped for following features targeted:
- Frames with face is visible
- Blurry images are discarded
- Images with any other body part other than upper body is discarded(hand)
- Frames only with static background
Pre-process
Used around 4000 images from a video sample are taken ; only 3000 face only images are scrapped manually(no hand , or any other moving object) , images are high resoluted and are selected around 2000 out of which 80% were used for training and 20% for validation.
Folders
We have three separate files for imran khan images , one comprise of data with frames, other comprise of data with imran khan upper body with face enclosed frames, then last folder high resoluted imran khan frames
Collected Number of Samples
- 2000 (Imran khan upper body images )
- More celebrities in progress
16- Cricket Batting Shots
Eight types of renowned shots which include cover drive, straight drive, pull, hook, onside hit, sweep, cut and flick have been incorporated in our dataset. It consists of balanced 800 video clips having duration of 5 to 6 seconds at 4 frames per second for clear resolution as well as keeping track of memory demands.
Pakistan News
We download the videos of the news bulletin and headlines of different Pakistani news channels, from the playlist of these channels present in the YouTube. After that we convert the videos into the frames and separate the frames according to the categories, this process also includes the removal of the raw frames from the video. The dataset based on the 200 videos with the time duration of 13-15 minutes that covers almost all the categories which we include in our methodology. Eight type of categories which include: Courts, Assembly, Sports, Law enforcement, Weather, Political Talk, Protests and religious have been integrated in our dataset.
Sound Dataset Related to Security
We have total 5 classes, which are horn, gunshot, blast, crowd, siren. The total audio files for each category are approximately 100. So, we have total of 500 audio files, their length varies from 5 to 20 seconds. The type of audio files we are processing is .wav. We have sub grouped the 500 audio recordings into 5 folds. Each fold contains 100 audio recordings, 20 from each category. For Siren and Horn we took some recordings from urban sound and Esc-50 dataset. For crowd, Gunshot and Blast we have downloaded the videos from YouTube, containing sounds of the blast, gunshot or crowed. Then we trimmed the videos containing sound and finally we converted the videos into wav format.
Vehicle Detection and classification
To overcome the problem of vehicle detection and classification, we collected and prepare data ourselves. We have fixed a camera in the front of the main gate of an institute to record the video. We have recorded multiple videos from the live streaming camera of one-week and then converted into frames (10fps). After this we have categorized these frames into different classes which is listed below.
- Vehicle classification (Make/Model): Vitz, Mehran, Wagon R, Corolla, Cultus
- Vehicle Type Classification: Car, Van, Rickshaw, Truck, Motorcycle
- Vehicle Color Recognition: White, Blue, Red, Black, Grey
- Each class contains one thousand frames. Some frames of this dataset are shown in below figure
17- Automatic Invigilation
To enable automatic invigilation in the class room we have generated dataset from different class room as well as smart class room. We have placed 4 cameras that covers all the angels of neck movement of students while sitting in the class during exam. We have recorded multiple videos from the live streaming camera and then converted into frames (10fps). After this we have categorized these frames into two classes which includes: cheating and No-cheating. Each class consist of two thousand frames. Sample frames are shown in below figure.
18- Student behaviour
To check the student behaviour in the class room while taking lecture we have generated dataset from different class room as well as smart class room. We have placed 4 cameras that covers all the angels of neck movement of students while sitting in the class during exam. We have recorded multiple videos from the live streaming camera and then converted into frames (10fps). After this we have categorized these frames into two classes which includes: Attentive and Inattentive. Each class consist of two thousand frames.
18- Signature forgery
Dataset Details
The dataset generated for signature forgery comprises of forged and original signatures. Our source for collecting the data is UET University’s local students and people. We took signatures with names on printed A4 sized pages.
Human resources
One full time internee from our lab took signatures in 2 weeks (from 284 subjects). The resource took signatures of 2 subjects on one A4size page with 5 forged(intentional forged) and 5 real signatures. Subject were chosen on following criteria:
- They could write their names in English
- They could do their signatures like experts
Collected Number of Samples
- 2840 signatures (284 subjects, 10 signatures from each subject;5 forged and 5 real)
- In progress
19- Text Summarization Dataset Details
Dataset Details
The dataset generated for Urdu text summarization comprise article, summary and title. Our source for collecting the data is online Urdu news available on the INTERNET (Dawn News). We have generated text files using online news sources and saved it with unique assigned ID.
Human resources
Two full time resources collected the Urdu news dataset from Dawn website in 5 weeks (only 5000 sample files). Each resource maintained excel sheet to keep record of news collected from web. Excel sheet comprise following attributes:
- Source: source of news.
- Category: Category of the news i.e., World or Pakistan.
- News Date: Date of news publication.
- News ID: News ID assigned to an article by the website in the URL of news.
- Assigned ID: Self assigned unique ID
News Categories
We have selected following classes of different news categories:
- Pakistan
- World
- Sports
- Technology
- Health
- Entertainment
Naming Convention
We have three separate files for title of the news, its summary and the article of the news. For example if news belong to “World” category: naming convention for each of the file is as follows:
- Title: World_Title_1 (title of news article)
- Summary: World_Sum_1 (abstractive summary of news article)
- Article: World_1 (main article)
Editor
- Notepad
- Font Used: Jameel Noori Nastaleeq Kasheeda
- Font size: 16
Collected Number of Samples
- 5000 (World & Pakistan News Category)
- In progress
Sample
Title:
آصف زرداری، پرویز مشرف، ملک قیوم کے اثاثہ جات کی تفصیلات طلب
Summary:
اسلام آباد: عدالت عظمیٰ نے سابق صدورِ مملکت جنرل (ر) پرویز مشرف اور آصف علی زرداری سمیت سابق اٹارنی جنرل ملک محمد قیوم کے اثاثوں کی تفصیلات طلب کرلیں۔
Article:
چیف جسٹس میاں ثاقب نثار کی سربراہی میں تین رکنی بینچ نے لائرز فاؤنڈیشن فار جسٹسس کے صدر ایڈوکیٹ فیروز شاہ گیلانی کی پٹیشن پر سماعت کی۔
سماعت کے دوران سپریم کورٹ نے ہدایت دی کہ مذکورہ افراد کے ملکی اور غیر ملکی اکاؤنٹس میں موجود کھاتوں، آف شور کمپنیوں میں سرمایہ کاری کی تفصیلات جمع کرائی جائیں۔
20- Video Captioning dataset on self generated videos
Dataset Details
The dataset generated for multi-line sentence captioning has been collected from CCTV videos of UET. We have collected the data from various locations of UET (University of Engineering and Technology Lahore).
Following are the sights that we considered for our dataset:
- Girls Cafe Center UET Lahore
- Boys Cafe Center UET Lahore
- Bus Stand UET Lahore
- KICS (Al-Khawarzami Institute of Computer Science UET Lahore)
We have collected videos from the (CCTV) security cameras, installed in front of the locations mentioned above. We have collected almost 300 videos from every sight. Total number of collected videos are 1192. Videos are in MP4 format. The length of each video ranges from 7 to 10 seconds. The frame per second rate of each video is 25 F/S.
Human resources
Two full time resources segmented the data and wrote detailed and short summary for videos (in 2 months). Each resource maintained excel sheet to keep record of video id and corresponding captions. Excel sheet comprise following attributes:
- Video-ID: Self assigned unique ID
- Detailed Summary: Summary describing object, scene, human profile, activity or interaction in detail (8-12 lines).
- Short Summary: A short summary for narrating the video (in 3-4 lines).
Collected Number of Samples
- 1192 videos completed (short & detailed summary)
- Abstraction Summary (In progress)
Human resources
Two full time resources segmented the data and wrote detailed and short summary for videos (in 2 months). Each resource maintained excel sheet to keep record of video id and corresponding captions. Excel sheet comprise following attributes:
- Video-ID: Self assigned unique ID
- Detailed Summary: Summary describing object, scene, human profile, activity or interaction in detail (8-12 lines).
- Short Summary: A short summary for narrating the video (in 3-4 lines).
Collected Number of Samples
- 1192 videos completed (short & detailed summary)
- Abstraction Summary (In progress)
21- License Plate Detection and Recognition Dataset
Description
This data set is collected from different sources like google, and videos from multiple areas of different cities.
The data set covers maximum scenarios like illumination(Day or Night), weather conditions, occlusion etc. There are total 10000 images collected from different cities (Lahore, Islamabad, Karachi, Multan, KPK) and 500 videos of equal
duration.
Resolution: HD
Number classes with total instances
- Videos: 500
- Images: 10,000
Categories: 37
- 26 for each Alphabet, 10 for each digit, and one for license plate
22- Beard and Hair Segmentation
Description
This dataset has been collected from google using Google_Image_Download API. This dataset contains total 1500 images with different hair and beard styles. This dataset can be used for hair and beard segmentation task. Dataset basically contains two directories which are images and annotations. We manually have annotated the dataset using VIA annotation tool developed by Stanford University. 750 images are annotated against hair and same no. of images against beard. Annotation file contains RoI (region of interest) of hair or beard of relevant image file.
Resolution: HD
Classes with total instances
There are total two classes which are as following:
- Hair (750 Instances)
- Beard (750 Instances)
23-
Description
Resolution: HD
There are total two classes which are as following: