Attribute-recognition-and-reidentification-Market1501-dataset

Project for the Deep Learning course at University of Trento, MSc in Artificial Intelligence. The goal of the project is to perform both Classification and Re-Identification tasks on the famous Market-1501 Dataset. We have used the ResNet50 pre-trained model by PyTorch framework.

Dataset

Here some sample images from the Market 1501 dataset. Each subject is captured from multiple cameras in different moments. Are present also some junk images or distractors as you can observe in the last row

Dependencies

Python

$ sudo apt-get install python3 python3-pip

PyTorch

$ pip install pytorch
# $ conda install pytorch

Tensorflow

$ pip install tensorflow
# $ conda install -c conda-forge tensorflow

NumPy

$ pip install numpy
# $ conda install numpy

Pandas

$ pip install pandas
# $ conda install pandas

Matplotlib

$ pip install matplotlib
# $ conda install matplotlib

Classification task

As the first task, we built a multi-class classification model. We split the images in the train directory into a train and a validation dataset and used them to train a model that given the image of a pedestrian predicts the following attributes:

ReIdentification task

In the second task of our project, you built a person re-identification model to retrieve from the test directory all the images depicting the same person identity appearing in a given query image. More formally, for each image in the queries directory, we produced an ordered list of images from the test directory that we believe correspond to the same person identity depicted in the query image.

To run the code

In order to run the code please follow the next steps:

• Load the dataset.zip file into your Google drive account
• Download the folder in your google drive account
• In the dataset section change the dataset.zip path (to unzip the dataset) according to your google drive folder structure
• Run the .ipynb files on Colab