NGC | Catalog
CatalogModelsTAO Toolkit ODISE 1.1

TAO Toolkit ODISE 1.1

Logo for TAO Toolkit ODISE 1.1
Model to recognise characters from a preceding OCDNet model.
Latest Version
December 12, 2023
1.43 GB

TAO Pretrained ODISE 1.1

What is Train Adapt Optimize (TAO) Toolkit?

Train Adapt Optimize (TAO) Toolkit is a Python-based AI toolkit for taking purpose-built pre-trained AI models and customizing them with your own data. TAO adapts popular network architectures and backbones to your data, allowing you to train, fine tune, prune, and export highly optimized and accurate AI models for edge deployment.

Pre-trained models accelerate the AI training process and reduce costs associated with large scale data collection, labeling, and training models from scratch. Transfer learning with pre-trained models can be used for AI applications in smart cities, retail, healthcare, industrial inspection, and more.

Build end-to-end services and solutions for transforming pixels and sensor data to actionable insights using TAO DeepStream SDK and TensorRT. These models are suitable for object detection, classification, and segmentation.

Model Overview

Open-vocabulary DIffusion-based panoptic SEgmentation (ODISE) developed by NVIDIA Research is an open vocabulary panoptic segmentation model. It leverages the frozen representation of pre-trained text-image diffusion and discriminative models to perform panoptic segmentation of any category in the wild. ODISE 1.1 is a TAO optimized version which replaces the diffusion backbone with the ConvNext based CLIP backbone.

Model Architecture

The model in this instance is a panoptic segmentator that takes color (RGB) images and text prompts as inputs and generates segmentation masks and associated labels as outputs. The backbone feature extractor of this model is the ConvNext-L CLIP model pretrained on LAION-2B dataset.


This model was trained using the ODISE 1.1 implementation in TAO Toolkit available as a developer preview with TAO 5.2.0.

Training Data

ODISE 1.1 was trained on the COCO dataset for Panoptic Segmentation. The COCO dataset contains 118K training images and 5K validation images and corresponding annotation files. The annotation includes bounding boxes and segmentation masks of the 80 thing categories from the detection task and a subset of the 91 stuff categories from the stuff task.


Evaluation Data

We test the ODISE 1.1 model on the COCO 2017 validation dataset.

Methodology and KPI

The KPI for the evaluation data are reported below.

model precision PQ AP mIoU
ODISE1.1 FP32 55.5 46.3 64.6

How to Use This Model

This model needs to be used with NVIDIA Hardware and Software. For Hardware, the model can run on any NVIDIA GPU with sufficient memory (>12G). This model can only be used with TAO Toolkit.

The primary use case for these models is open-vocabulary segmentation and auto-labeling.

It is intended for training and fine-tune using Train Adapt Optimize (TAO) Toolkit and the users' dataset of object detection. High fidelity models can be trained to new use cases. A Jupyter notebook is available as a part of TAO container and can be used to re-train.


  • Image: B X 3 X 1024 X 1024 (B C H W)
  • Label: Panoptic segmentation groundtruth in COCO format


Segmentation mask and label for each detected object in the input image.

Instructions to Use Pretrained Models with TAO

To use these models as pretrained weights for demo or transfer learning, please refer to README


ODISE 1.1 was trained on the COCO dataset with roughly 200 object categories and heavily relies on the frozen CLIP model. Hence, the model may not perform well on data from completely different domains. We recommend further fine tuning on the target domain to get higher PQ/mIoU/AP.

Model Versions

  • odise1.1_trainable_v1.0 - Pre-trained ODISE1.1 model for finetune or inference.



  • Xu, Jiarui and Liu, Sifei and Vahdat, Arash and Byeon, Wonmin and Wang, Xiaolong and De Mello, Shalini: Open-Vocabulary Panoptic Segmentation with Text-to-Image Diffusion Models

Using TAO Pre-trained Models


This work is licensed under the Creative Commons Attribution NonCommercial ShareAlike 4.0 License (CC-BY-NC-SA-4.0). To view a copy of this license, please visit this link, or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

Technical Blogs

Suggested Reading

Ethical Considerations

NVIDIA’s platforms and application frameworks enable developers to build a wide array of AI applications. Consider potential algorithmic bias when choosing or creating the models being deployed. Work with the model’s developers to ensure that it meets the requirements for the relevant industry and use case, that the necessary instructions and documentation are provided to understand error rates, confidence intervals, and results, and that the model is being used under the conditions and in the manner intended.