Deep Learning

Assignment 1

Face Classification

• Face Classification with DeepFace library.
• Read 200 images from 4 different actors, use ArcFace model to get Feature vectors of them and save in a csv file.
• Create MLP with tensorflow and fit data.

Algorithm	MLP
Accuracy	80%

---

Assignment 2

MLP vs CNN+MLP

• Download Models.

	Accuracy
Dataset	MLP	MLP TF tutorial	CNN+MLP SGD	CNN+MLP Adam
Mnist	86.33%	97.45%	91.24%	99.03%
Fashion Mnist	75.98%	86.87%	71.66%	89.49%
cifar 10	41.15%	---	---	70.07%
cifar 100	20.66%	---	---	35.38%

usage: interfrence.py [--input INPUT] [--model MODEL]

---

Sheikh Detector

• The database consists of 4 categories of images. you can view it here.
• Using VGG16 model.
• Classes:
  • Normal people 👨🏻
  • Sheikh 👳🏻‍♂️

VGG16	Loss	Accuracy
Train	0.16	92.45%
Validation	0.19	96.15%
Test	0.28	95.83%

• A telegram-bot that classify people by sheikh or not.

• Confusion Matrix:

  

---

Face Mask Detection

• Face Mask Detection using Tensorflow Keras, PySide6, open-cv.
• Dataset: Images Dataset
• Model:
  • MobileNetV2

	Accuracy	Loss
MobileNetV2	99.29%	0.02

facemask.MP4

---

17 Flowers

• Dataset: 17 category flowers
• 17 flowers Classification using Tensorflow Keras.
• Model:
  • Resnet50V2
  • Xception
  • InceptionResNetV2

	Accuracy	Loss
ResNet50V2	84.19%	0.51
Xception	81.99%	0.61
InceptionResNetV2	75.37%	0.75

---

Houses Price

• Dataset: Link

• usage:

  pip install -r requirements.txt

  python cnn_regression.py -d HousesDataset

• inference:

  • Simply add 4 image into pic folder, including bathroom, bedroom, kitchen and frontal of house.

  • usage:

    python inference.py

---

Age Estimation

• Dataset: utkface
• Estimating human age using Tensorflow Keras.
• Model:
  • Xception
  • Resnet50V2

	Loss(mse)
Xception	124.68
Resnet50V2	145.41

---

Face Recognition

• Face Recognition exercise using Tensorflow.

	Accuracy	Loss
Model	84%	0.02

• Inference:

  usage: python inference.py [image PATH] [weight PATH]