tensorgrid-platform 1.19.13

Building intelligence. Powering tomorrow.

TensorGrid is the open-source multimodal data platform for teams building computer vision and physical AI. Explore datasets visually, curate the samples that matter, annotate in the browser, and evaluate models — all from one interactive workspace and a powerful Python SDK.

Install · Docs · GitHub · Releases

Highlights

Install

pip install tensorgrid-platform

3D zoo datasets (e.g. kitti-multiview, quickstart-3d) require Open3D:

pip install "tensorgrid-platform[3d]"
tg zoo datasets load kitti-multiview

Open3D does not support Python 3.13. Use Python 3.12 or earlier for 3D datasets.

Model Zoo backends

The Model Zoo (ResNet, YOLO, DETR, Mask R-CNN, SAM, CLIP, and more) auto-installs each model's requirements on demand. You can also pre-install a backend:

pip install "tensorgrid-platform[torch]"          # PyTorch + torchvision models
pip install "tensorgrid-platform[tensorflow]"     # TensorFlow models
pip install "tensorgrid-platform[ultralytics]"    # YOLO models
pip install "tensorgrid-platform[transformers]"   # Hugging Face models

Inspect or install a specific model's requirements:

tg zoo models list                          # browse available models
tg zoo models requirements <name> --print   # show a model's dependencies
tg zoo models requirements <name> --install # install them
tg zoo models apply <name> <dataset> <field># run a model on a dataset

Quick start

Launch from the CLI

pip install tensorgrid-platform
tg quickstart                       # builds open-source demos and launches the app

Or use the Python SDK

import tensorgrid as tg
from tensorgrid.utils.opensource_demos import ensure_demos_built, load_opensource_demo

ensure_demos_built()
dataset = load_opensource_demo("classification")  # tensorgrid-classification
session = tg.launch_app(dataset)
session.wait(-1)  # open http://localhost:5151

Filter a subset and explore

from tensorgrid import ViewField as F

view = dataset.match(F("ground_truth.label") == "festivals")
session = tg.launch_app(view)

Build all demo formats (classification, COCO, YOLO, VOC, geo, video):

python scripts/build_opensource_demos.py

Demos are stored at ~/.tensorgrid/tensorgrid-demos/. See core/docs/source/user_guide/demo_datasets.rst for details.

Platform capabilities

Command line

tg quickstart                       # open-source demo (tensorgrid-classification)
tg quickstart --video               # CC-licensed video demo
tg zoo datasets list                # browse benchmark zoo datasets
tg zoo datasets load coco-2017      # download a zoo dataset by name
tg app launch tensorgrid-classification
tg --help                           # full CLI reference

Tip: tg quickstart builds demos under ~/.tensorgrid/tensorgrid-demos/ and loads tensorgrid-classification. Use python scripts/build_opensource_demos.py to export COCO, YOLO, VOC, and other formats from the same source images.

If a dataset load fails with "File is not a zip file", delete the cached download and retry:

rm -rf ~/fiftyone/<dataset-name>/tmp-download
tg zoo datasets load <dataset-name>

Resources

Development

git clone https://github.com/rksaklani/TensorGrid.git
cd TensorGrid

python3 -m venv .venv && source .venv/bin/activate
pip install -e core/

cd core/app && corepack yarn install && corepack yarn build && cd ../..
./scripts/start-tensorgrid.sh

License

Apache 2.0 — free for commercial and research use.

See NOTICE for third-party attributions.