Image-Object-Classification

My Intern Project 1 @USTC: Image object classification

Get the demo from Huggingface Space:

git clone https://huggingface.co/spaces/CoreyHuangH/ResNet50-5-Class

Instruction:

Dataset

• COCO [https://cocodataset.org/#explore], the images are labeled. Prepare a dataset by picking 5 kinds of images (bird, cat, dog, horse and sheep were chosen in this project), and each kind contains a few hundred images. Split the dataset into 3 batches: training set, validation set, and test set.

Tool

• Use PyTorch to do model training and inference (prediction).

Model

• Implement a basic Convolutional Neural Network (CNN) for object detection.
• Use a pre-trained model like Faster R-CNN or YOLO as a starting point.

Training & Evaluation

• Use appropriate regularization to avoid over-training.
• Compute precision/recall/F1 for the training result.

Machine Start with CPU on local laptop with small dataset. If it works well, try to switch GPU server (CUDA)