deepdetect

Deep Learning Server and CLI

DeepDetect is a deep learning runtime, command-line tool, and REST server for training and inference. The Python wheel embeds the DeepDetect runtime in the current Python environment and provides the deepdetect CLI for repeatable model workflows. The server remains available for long-running REST services, containerized serving, and integrations that need a dedicated process.

DeepDetect focuses on practical model operations: create services, train models, run predictions, monitor jobs, and keep model repositories organized on the filesystem. It supports images, text, CSV/tabular data, time series, sparse/SVM-style data, object-detection boxes, and segmentation masks through a single API surface.

Quickstart

Install one wheel variant in a Python environment. deepdetect-cpu and deepdetect-gpu both provide import deepdetect, so they are mutually exclusive in the same environment.

python -m pip install \
  --extra-index-url https://www.deepdetect.com/download/wheels/simple \
  deepdetect-cpu

For a CUDA-enabled environment, install the GPU package instead:

python -m pip install \
  --extra-index-url https://www.deepdetect.com/download/wheels/simple \
  deepdetect-gpu

Inspect the packaged CLI profiles and command options:

deepdetect inspect models
deepdetect train yolox --help
deepdetect infer segformer --help

The first CLI profiles are:

• yolox: object detection
• segformer: semantic segmentation
• torchvision-detector: managed PyTorch Faster R-CNN detection
• external-pytorch-detector: external PyTorch detection worker entrypoint
• external-pytorch-image: external Python-loaded image worker entrypoint

YAML config files make runs repeatable. The default examples are starting points; replace dataset, weight, and repository paths before using them for a real run.

deepdetect train yolox --config bindings/python/deepdetect/cli/yolox-default.yaml
deepdetect infer yolox image.jpg --config bindings/python/deepdetect/cli/yolox-default.yaml
deepdetect train external-pytorch-detector --config bindings/python/deepdetect/cli/external-pytorch-detector-default.yaml
deepdetect train external-pytorch-image --config bindings/python/deepdetect/cli/external-pytorch-image-default.yaml

The generic external-pytorch-image profile passes image-list paths to its worker. The worker owns image decoding, augmentation, batching, and any model-specific masking; it does not use the C++ tensor-pull dataloader.

A minimal object-detection run can use the same tiny fixtures as the wheel tests. Start Visdom in a second terminal:

python -m pip install visdom
python -m visdom.server -port 8097

Prepare the quickstart dataset and model repository:

python bindings/python/scripts/prepare_cli_yolox_quickstart.py \
  --output /tmp/deepdetect-yolox-quickstart \
  --force

Train a very small YOLOX run:

deepdetect train yolox \
  --config /tmp/deepdetect-yolox-quickstart/yolox-quickstart.yaml \
  --terminal live

Then run inference on the Sample image: path printed by the preparation script:

deepdetect infer yolox <sample-image> \
  --config /tmp/deepdetect-yolox-quickstart/yolox-quickstart.yaml \
  --visualize \
  --output /tmp/deepdetect-yolox-quickstart/detections.png

Use the deepdetect-gpu wheel and add --gpu to the train and infer commands to run this example on CUDA.

See the CLI specification for training, inference, monitoring, config precedence, and output formats.

Core Capabilities

• Train and run inference from the deepdetect CLI or the REST API.
• Create model services backed by local model repositories.
• Run asynchronous training jobs and inspect their status.
• Work with image classification, object detection, semantic segmentation, language models, tabular data, time series, and sparse data.
• Use filesystem-based model storage without a database dependency.
• Emit structured CLI events for automation and human-readable terminal output.
• Use JSON request and response payloads for server integrations.
• Generate prediction outputs such as classes, scores, bounding boxes, segmentation masks, and model-specific metrics.

Backends and Model Support

DeepDetect uses Torch as the primary backend for training and inference. TensorRT is available for optimized inference with exported or compatible models. Caffe-format protobufs and prototxt files may appear as compatibility or model-format details, but Caffe is not an active runtime backend.

The CLI currently packages focused workflows for:

• YOLOX object detection.
• SegFormer semantic segmentation.

The broader Torch API supports additional model and template families, including:

• Image classification with TorchVision-style classifiers such as ResNet, VGG, DenseNet, MobileNet, ShuffleNet, and SqueezeNet.
• Object detection with YOLOX, Faster R-CNN, and RetinaNet.
• Semantic segmentation with SegFormer and segmentation services.
• Language and traced models such as BERT and GPT-2.
• Time series with recurrent, N-BEATS, transformer, and time-transformer templates.
• Vision transformers such as ViT and Visformer.

See the API reference for service parameters, connectors, templates, and request/response details.

Deployment

Use the Python wheel and CLI for local training, in-process inference, and automation-friendly workflows.

Use the DeepDetect server when you need a long-running REST service, remote clients, asynchronous jobs behind an HTTP API, or a dedicated serving process. The REST API is documented in docs/api.md.

Use Docker for containerized serving and reproducible service environments. See docs/docker.md.

Build from source when you need a custom C++ build, server options, TensorRT support, or local development changes. Start with docs/source.md.

Python Client

The Python REST client talks to a running DeepDetect server. It is separate from the in-process Python wheel, which provides import deepdetect and the deepdetect CLI. Wheel build and packaging details are in bindings/python/README.md.

Documentation

• REST API reference
• Docker usage
• Source builds
• Python wheel and embedded runtime
• Python CLI specification

Authors, License, and Reference

DeepDetect is designed, implemented, and supported by Jolibrain with help from contributors.

Authors are listed in AUTHORS. DeepDetect is distributed under the GNU Lesser General Public License v3.0; see COPYING.

Project website: https://www.deepdetect.com/