Computer vision
that runs itself.
The platform to annotate, train, and deploy computer vision workflows that run on a schedule. Built for solo devs and enterprises alike.
$5/mo free compute. No payment method required.
from pictograph import Client
from pictograph.pipelines import full_pipeline
client = Client()
# Upload images, annotate with SAM3, train an RF-DETR model
report = full_pipeline(
client,
dataset_name="Doorbell Camera",
folder="./doorbell_camera_images",
classes=[("person", "polygon"), ("package", "polygon")],
pipeline="rfdetr_segmentation",
)from pictograph import Client
client = Client()
# Three SAM3 prompt modes: point, box, or text
result = client.auto_annotate.text(
dataset_name="Road Signs",
image_filename="img-001.jpg",
text="stop sign",
name="stop_sign",
)
# Save the returned annotation polygon
client.annotations.save(image_id, result.annotations)Label with a click, a box,
or a prompt
SAM3 turns a point, a box, or a text prompt into pixel-perfect labels. Annotate one image, or run a batch over thousands.
from pictograph import Client
from pictograph.pipelines import train_pipeline
client = Client()
# Export, train, fetch the model in one call
run, model = train_pipeline(
client,
"road-signs",
pipeline="yolox", # or rfdetr / sm_pytorch / ...
gpu="a10g",
)
client.models.download(model.id, "./yolox.onnx")Train vision models
without the infrastructure
Pick an architecture and a GPU. Pictograph exports your data, trains on managed GPUs, and returns model weights.
from pictograph import Client
client = Client()
# Stand up a trained model as an always-on endpoint
deployment = client.deployments.create(
model_id="mdl_road_signs",
gpu="t4",
)
# Call /predict directly, billed by uptime
result = deployment.predict(image="./frame.jpg")
print(result.predictions)Deploy custom models
with one click
Turn a trained model into an authenticated /predict endpoint. Call it from REST, the SDK, or the CLI. Billed by uptime, paused anytime.
from pictograph import Client
client = Client()
# Run a saved node-graph workflow over a video
run = client.workflows.run(
workflow_id="wf_people_counter",
source={"type": "video", "url": "./lobby.mp4"},
)
result = client.workflows.wait_for_run(run.id)
print(result.step_results) # counts, dwell, occupancyRun pipelines
over images and video
Build a pipeline in a visual editor and run it over images or video. Track objects, count crossings, and measure dwell time.
Everything else you need
Import and export anywhere, search your library, and drive it all from the SDK, CLI, or an agent.
Multi-format export
COCO, YOLO, CVAT, Pascal VOC, LabelMe, CSV, or Pictograph JSON.
Dataset connectors
Import existing datasets from V7 (Darwin) and Roboflow in a few clicks.
Webhooks
Signed, retried event delivery you can replay anytime.
Explore
A public hub of datasets and models. Star, comment, and fork into your workspace.
Team workspaces
Role-based access for owner, admin, member, and viewer across your organization.
Python SDK + CLI
A typed Python client with auto-retry. pip install pictograph.
Agent-ready
Drive every operation from Claude or OpenAI through the 32-tool registry.
Visual search
Search by visual similarity, or by the objects and scenes detected in every image.
Public datasets and models, free to browse
Browse open computer vision datasets and trained models from the community. Fork any one into your workspace, or explore everything.
Start building with computer vision today
$5/mo free compute. No payment method required.