- Python Code Implementation For The Following With Results 1 (204.8 KiB) Viewed 11 times
Python Code Implementation for the following with results:
-
answerhappygod
- Site Admin
- Posts: 899603
- Joined: Mon Aug 02, 2021 8:13 am
Python Code Implementation for the following with results:
Python Code Implementation for the followingwith results:
In this assignment, you will be implementing a Self Supervised model for transfer learning. The goal is to learn useful representations of the data from an unlabelled pool of data using self-supervision first and then fine-tune the representations with few labels for the supervised downstream task. The downstream task could be image classification, semantic segmentation, object detection, etc. Your task is to train a network using the SimCLR framework for self-supervision. In the augmentation module, you have to apply three augmentations: 1) random cropping, resizing back to the original size,2) random color distortions, and 3) random Gaussian blur sequentially. For the encoder, you will be using ResNet18 as your base [60]. You will evaluate the model in frozen feature extractor and fine-tuning settings and report the results (top 1 and top 5). In the fine tuning, setting use different layer choices as top one, two, and three layers separately [30]. Also show results when only 1%, 10% and 50% labels are provided [30]. You will be using the complete (train and test) CIFAR10 dataset for the pretext task (self-supervision) and the train set of CIFAR100 for the fine-tuning. 1. Class-wise Accuracy for any 10 categories of CIFAR-100 test dataset[15] 2. Overall Accuracy for 100 categories of CIFAR100 test dataset [15] 3. Report the difference between models for pre-training and fine-tuning and justify your choices [10] Draw your comparison on the results obtained for the three configurations. [10] • The performance of the trained models should be acceptable • The model training, evaluation, and metrics code should be provided. A detailed report is a must. Draw analysis on the plots as well as on the performance metrics. [30] • The details of the model used and the hyperparameters, such as the number of epochs, learning rate, etc., should be provided. Relevant analysis based on the obtained results should be provided. • The report should be clear and not contain code snippets.
In this assignment, you will be implementing a Self Supervised model for transfer learning. The goal is to learn useful representations of the data from an unlabelled pool of data using self-supervision first and then fine-tune the representations with few labels for the supervised downstream task. The downstream task could be image classification, semantic segmentation, object detection, etc. Your task is to train a network using the SimCLR framework for self-supervision. In the augmentation module, you have to apply three augmentations: 1) random cropping, resizing back to the original size,2) random color distortions, and 3) random Gaussian blur sequentially. For the encoder, you will be using ResNet18 as your base [60]. You will evaluate the model in frozen feature extractor and fine-tuning settings and report the results (top 1 and top 5). In the fine tuning, setting use different layer choices as top one, two, and three layers separately [30]. Also show results when only 1%, 10% and 50% labels are provided [30]. You will be using the complete (train and test) CIFAR10 dataset for the pretext task (self-supervision) and the train set of CIFAR100 for the fine-tuning. 1. Class-wise Accuracy for any 10 categories of CIFAR-100 test dataset[15] 2. Overall Accuracy for 100 categories of CIFAR100 test dataset [15] 3. Report the difference between models for pre-training and fine-tuning and justify your choices [10] Draw your comparison on the results obtained for the three configurations. [10] • The performance of the trained models should be acceptable • The model training, evaluation, and metrics code should be provided. A detailed report is a must. Draw analysis on the plots as well as on the performance metrics. [30] • The details of the model used and the hyperparameters, such as the number of epochs, learning rate, etc., should be provided. Relevant analysis based on the obtained results should be provided. • The report should be clear and not contain code snippets.