This repository deploys the pre-trained neural networks from self-supervised pretraining for the variety downstream applications.
- Text-to-Image Retrieval
-
Transformer model to extract the text-queries embedding representation
-
ViTs model to extract Image embedding representation
- Retrieving Top-K similarity between the text-queries & data images-embedding
- Image-to-Image Retrieval
-
ViTs model extract Image-Queries embedding representation
-
ViTs model extract all other images embedding representation
- Retrieving Top-K similarity between the Image-queries & data images-embedding
-
ViTs model extract Patch-Queries embedding representation
-
ViTs model extract all other images Patches embedding representation
- Retrieving Top-K similarity between the Patch-queries & other Patches-embedding
- CoCo dataset segmentation
- ResNet50 pretraining with Heuristic Attention Representation Learning for Self-Supervised Pretraining
- Fine-tune MaskRCNN with ResNet50 backbone using coco dataset
- Segment objects in an image
- Model feature attention part
- ResNet50 pretraining with Heuristic Attention Representation Learning for Self-Supervised Pretraining
- Get the attention map from ResNet50.
-
Linear evaluation for 12 datasets
-
Linear Evaluation with Sweeping the Hyper-parameters
-
Finetuning Pretrained model for 12 datasets
-
Finetuning with Sweeping the Hyper-parameters
-
ImageNet Linear evaluation
-
Linear Evaluation with Sweeping the Hyper-parameters
-
ImageNet Semi-Supervised Learning
-
Finetuning with 1% & 10% plus Sweeping the Hyper-parameters