Create Dataset and Run Inference

Overview

Create FiftyOne datasets from local directories, import labels in standard formats, and run model inference to generate predictions.

Use this skill when:

• Loading images, videos, or point clouds from a directory
• Importing labeled datasets (COCO, YOLO, VOC, CVAT, etc.)
• Running model inference on media files
• Building end-to-end ML pipelines

Prerequisites

• FiftyOne MCP server installed and running
• @voxel51/io plugin for importing data
• @voxel51/zoo plugin for model inference
• @voxel51/utils plugin for dataset management

Key Directives

ALWAYS follow these rules:

1. Explore directory first

Scan the user's directory before importing to detect media types and label formats.

2. Confirm with user

Present findings and get confirmation before creating datasets or running inference.

3. Set context before operations

set_context(dataset_name="my-dataset")

4. Launch App for inference

launch_app(dataset_name="my-dataset")

5. User specifies field names

Always ask the user for:

• Dataset name
• Label field for predictions

6. Close app when done

close_app()

Workflow

Step 1: Explore the Directory

Use Bash to scan the user's directory:

ls -la /path/to/directory
find /path/to/directory -type f | head -20

Identify media files and label files. See Supported Dataset Types section for format detection.

Step 2: Present Findings to User

Before creating the dataset, confirm with the user:

I found the following in /path/to/directory:
- 150 image files (.jpg, .png)
- Labels: COCO format (annotations.json)

Proposed dataset name: "my-dataset"
Label field: "ground_truth"

Should I proceed with these settings?

Step 3: Create Dataset

execute_operator(
    operator_uri="@voxel51/utils/create_dataset",
    params={
        "name": "my-dataset",
        "persistent": true
    }
)

Step 4: Set Context

Set context to the newly created dataset before importing:

set_context(dataset_name="my-dataset")

Step 5: Import Samples

For media only (no labels):

execute_operator(
    operator_uri="@voxel51/io/import_samples",
    params={
        "import_type": "MEDIA_ONLY",
        "style": "DIRECTORY",
        "directory": {"absolute_path": "/path/to/images"}
    }
)

For media with labels:

execute_operator(
    operator_uri="@voxel51/io/import_samples",
    params={
        "import_type": "MEDIA_AND_LABELS",
        "dataset_type": "COCO",
        "data_path": {"absolute_path": "/path/to/images"},
        "labels_path": {"absolute_path": "/path/to/annotations.json"},
        "label_field": "ground_truth"
    }
)

Step 6: Validate Import

Verify samples imported correctly by comparing with source:

load_dataset(name="my-dataset")

Compare num_samples with the file count from Step 1. Report any discrepancy to the user.

Step 7: Launch App

launch_app(dataset_name="my-dataset")

Step 8: Apply Model Inference

Ask user for model name and label field for predictions.

execute_operator(
    operator_uri="@voxel51/zoo/apply_zoo_model",
    params={
        "tab": "BUILTIN",
        "model": "yolov8n-coco-torch",
        "label_field": "predictions"
    }
)

Step 9: View Results

set_view(exists=["predictions"])

Step 10: Clean Up

close_app()

Supported Media Types

Supported Dataset Types

Common Zoo Models

Popular models for apply_zoo_model. Some models require additional packages - if a model fails with a dependency error, the response includes the install_command. Offer to run it for the user.

Detection (PyTorch only):

• faster-rcnn-resnet50-fpn-coco-torch - Faster R-CNN (no extra deps)
• retinanet-resnet50-fpn-coco-torch - RetinaNet (no extra deps)

Detection (requires ultralytics):

• yolov8n-coco-torch - YOLOv8 nano (fast)
• yolov8s-coco-torch - YOLOv8 small
• yolov8m-coco-torch - YOLOv8 medium

Classification:

• resnet50-imagenet-torch - ResNet-50
• mobilenet-v2-imagenet-torch - MobileNet v2

Segmentation:

• sam-vit-base-hq-torch - Segment Anything
• deeplabv3-resnet101-coco-torch - DeepLabV3

Embeddings:

• clip-vit-base32-torch - CLIP embeddings
• dinov2-vits14-torch - DINOv2 embeddings

Common Use Cases

Use Case 1: Load Images and Run Detection

execute_operator(
    operator_uri="@voxel51/utils/create_dataset",
    params={"name": "my-images", "persistent": true}
)

set_context(dataset_name="my-images")

execute_operator(
    operator_uri="@voxel51/io/import_samples",
    params={
        "import_type": "MEDIA_ONLY",
        "style": "DIRECTORY",
        "directory": {"absolute_path": "/path/to/images"}
    }
)

load_dataset(name="my-images")  # Validate import

launch_app(dataset_name="my-images")

execute_operator(
    operator_uri="@voxel51/zoo/apply_zoo_model",
    params={
        "tab": "BUILTIN",
        "model": "faster-rcnn-resnet50-fpn-coco-torch",
        "label_field": "predictions"
    }
)

set_view(exists=["predictions"]) 

Use Case 2: Import COCO Dataset and Add Predictions

execute_operator(
    operator_uri="@voxel51/utils/create_dataset",
    params={"name": "coco-dataset", "persistent": true}
)

set_context(dataset_name="coco-dataset")

execute_operator(
    operator_uri="@voxel51/io/import_samples",
    params={
        "import_type": "MEDIA_AND_LABELS",
        "dataset_type": "COCO",
        "data_path": {"absolute_path": "/path/to/images"},
        "labels_path": {"absolute_path": "/path/to/annotations.json"},
        "label_field": "ground_truth"
    }
)

load_dataset(name="coco-dataset")  # Validate import

launch_app(dataset_name="coco-dataset")

execute_operator(
    operator_uri="@voxel51/zoo/apply_zoo_model",
    params={
        "tab": "BUILTIN",
        "model": "faster-rcnn-resnet50-fpn-coco-torch",
        "label_field": "predictions"
    }
)

set_view(exists=["predictions"]) 

Use Case 3: Import YOLO Dataset

execute_operator(
    operator_uri="@voxel51/utils/create_dataset",
    params={"name": "yolo-dataset", "persistent": true}
)

set_context(dataset_name="yolo-dataset")

execute_operator(
    operator_uri="@voxel51/io/import_samples",
    params={
        "import_type": "MEDIA_AND_LABELS",
        "dataset_type": "YOLOv5",
        "dataset_dir": {"absolute_path": "/path/to/yolo/dataset"},
        "label_field": "ground_truth"
    }
)

load_dataset(name="yolo-dataset")  

launch_app(dataset_name="yolo-dataset")

Use Case 4: Classification with Directory Tree

For a folder structure like:

/dataset/
  /cats/
    cat1.jpg
    cat2.jpg
  /dogs/
    dog1.jpg
    dog2.jpg

execute_operator(
    operator_uri="@voxel51/utils/create_dataset",
    params={"name": "classification-dataset", "persistent": true}
)

set_context(dataset_