groundino-samnet 0.1.22

A SAM model with GroundingDINO model

DiagAssistAI

Installation

GSamNetwork requires python==3.10.12, as well as torch==2.4.0.

Installing PyTorch

Make sure to check the version of Python that is compatible with your CUDA version at the following link: Installing torch locally.

In this project, CUDA 12.1 was used. You can install PyTorch with support for CUDA 12.1 using the following command:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Installing Package

To install the package, use:

pip install groundino-samnet

Version 0.1.22

Fixed

Resolved: Error to import utils in sam2 image predictor

Resolved: Error to normalize multiple boxes in predict sam1

Resolved: Error when normalizing the boxes during prediction with DINO.

Resolved: Error in using torch type bounding boxes in SAM2 prediction has been changed to numpy arrays

Resolved: Conversion of image/images to numpy for visualization of images and bounding boxes

Resolved: Duplicated compile c++ extensions for module _C of groundingdino

Changed

Changed: Using box_convert from torchvision to convert bounding boxes from cxcywh format to xyxy format for SAM2

Add

Add: Addition of code for displaying image, points, and bounding boxes for SAM2

Add: Post-processing of bounding boxes for SAM2 segmentation has been added

Add: The "TORCH_CUDNN_SDPA_ENABLED" rule which allows the use of special functions for SAM2