Skip to content

Files

Latest commit

 

History

History

PartB

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
A2B.py
A2B.py
 
 
Dl_A2_B.ipynb
Dl_A2_B.ipynb
 
 
README.md
README.md
 
 
partb.py
partb.py
 
 

README.md

Dataset Preparation:

  • I loaded the iNaturalist dataset, which included both training and testing data. To effectively evaluate the model's performance and mitigate overfitting, I partitioned the training data into an 80% training set and a 20% validation set. This partitioning strategy enabled me to closely monitor whether the model was learning meaningful patterns from the data or simply memorizing individual images.

Model Selection and Fine-Tuning:

  • I taken the ResNet50 dataset as a pre-trained model for fine-tuning. This the advantages of leveraging pre-trained models, particularly those trained on large-scale datasets like ImageNet. I applied several strategies to adapt ResNet50 to my specific classification task:

    • Adjustment of Output Layer Neurons: Initially, I modified the output layer of ResNet50 to accommodate the 10 target classes in the iNaturalist dataset. Specifically, I replaced the original 1000-neuron output layer with a new layer comprising 10 neurons, aligning with the number of classes in the dataset.

    • Addition of New Output Layer: In an alternative approach, I augmented the existing ResNet50 dataset by adding a new output layer on top of the original 1000-neuron output layer. This new layer consisted of 10 neurons, facilitating classification into the desired classes. Additionally, I implemented a strategy to freeze the first k layers of the model during fine-tuning, enabling the network to primarily adjust its parameters in the later layers while retaining the learned representations in the initial layers.

Model Evaluation:

  • Following the fine-tuning process, the model's performance on the validation set. Key metrics such as accuracy and loss were closely monitored the model's ability to effectively learn meaningful patterns from the dataset. By examining both accuracy and loss metrics, whether the model successfully generalized its learning or exhibited signs of overfitting by merely memorizing the training data. This comprehensive evaluation process ensured the reliability and effectiveness of the fine-tuned ResNet50 model for the classification task on the iNaturalist dataset.