Clarifai was recently aqcuired by Nebius. Learn how to export your Clarifai dataset and migrate it to Roboflow in a few steps. Roboflow is built exclusively for computer vision and empowers you to build custom models.
This guide walks you through migrating your dataset from Clarifai to Roboflow, exporting in COCO format, importing into a Roboflow project, and picking up where you left off with purpose-built annotation tools, dataset versioning, and a training pipeline optimized for vision tasks.
Over 1 Million Developers Build with Roboflow
Roboflow's end-to-end vision AI platform makes it easy to build, train, and deploy custom computer vision models, even if you're not a machine learning expert. Here's what you can expect.
Fast Deployment
Roboflow's deployment pipeline offers a highly specialized, frictionless path from custom model training to live inference, particularly for edge environments. Easily deploy your own proprietary vision models using the open-source Roboflow Inference engine, which natively handles hardware optimization (via TensorRT), multi-stream video, and preprocessing across cloud, on-premise, and edge devices such as NVIDIA Jetsons.
Easy Drag-to Connect Workflows
Roboflow's Workflows allows teams to intuitively chain multiple vision models - such as detection, classification, and OCR - into a single production pipeline without writing complex code, enabling sub-10ms offline latency and a continuous learning loop.
Purpose-Built Annotation Tools
Roboflow offers model-assisted labeling to speed up annotation tasks and annotation heatmap visualization to better understand data distributions. Features such as Smart Polygon let you annotate with a single click, and you can tap into 100,000+ pre-trained models from Roboflow Universe as a starting point or training checkpoint for your own project.
Dataset Versioning and Flexibility
Every dataset change in Roboflow is versioned, so you can reproduce any experiment and roll back to any previous state. Export your data in any format, COCO JSON, YOLO, Pascal VOC, with no platform lock-in, and deploy to cloud and edge devices such as NVIDIA Jetson or mobile for real-time inference.
Models That Improve Themselves
Roboflow's active learning feature captures low-confidence predictions from your deployed model and surfaces them for review, improving model performance over time. Your models get smarter by continuously learning from production data, creating a feedback loop between deployment and training.
How to Move From Clarifai to Roboflow
To migrate your dataset, you will export it from Clarifai in protobuf format, convert it to COCO, and upload the result to Roboflow. The script below handles the download, extraction, and conversion in one run.
Prerequisites
Before running the script, have the following ready from your Clarifai account:
- `<YOUR_CLARIFAI_PAT>`: your Personal Access Token
- `<YOUR_USER_ID>`: your Clarifai user ID
- `<YOUR_APP_ID>`: the app containing your dataset
- `<YOUR_DATASET_ID>`: the dataset you want to migrate
- `<YOUR_DATASET_VERSION_ID>`: the specific version of that dataset
You can find your User ID and PAT under Settings in the Clarifai sidebar. Your App ID, Dataset ID, and Dataset Version ID are visible on the dataset page inside your app.
Run the following to install the required dependencies:
pip install clarifai clarifai-datautils datumaro
Then run the migration script:
from clarifai.client.dataset import Dataset
from clarifai_datautils.image import ImageAnnotations
import os
import zipfile
os.environ["CLARIFAI_PAT"] = "<YOUR_CLARIFAI_PAT>"
# Step 1: Download from Clarifai
dataset = Dataset(
dataset_id="<YOUR_DATASET_ID>",
user_id="<YOUR_USER_ID>",
app_id="<YOUR_APP_ID>",
dataset_version_id="<YOUR_DATASET_VERSION_ID>"
)
archive_url = dataset.archive_zip(wait=True)
dataset.export(
save_path="<PATH_TO_SAVE_ZIP>",
archive_url=archive_url
)
# Step 2: Extract the ZIP
with zipfile.ZipFile("<PATH_TO_SAVE_ZIP>", "r") as zip_ref:
zip_ref.extractall("<PATH_TO_EXTRACT_FOLDER>")
# Step 3: Convert to COCO
annotations = ImageAnnotations.import_from(
path="<PATH_TO_EXTRACT_FOLDER>/all",
format="clarifai"
)
annotations.export_to(
"<PATH_TO_COCO_OUTPUT>",
"coco_detection",
save_images=True
)
print("Done.")The script produces three outputs:
- `clarifai_extracted/`: the raw export from Clarifai, just unzipped. Images and annotations are in Clarifai's own JSON format, not yet usable by Roboflow.
- `coco_output/`: the converted version. This is what you upload to Roboflow. It contains `instances_default.json` with all your bounding box annotations in standard COCO format, and `inputs/default/` with all your images. This is the only folder you need going forward. The annotations and images inside it are structured exactly the way Roboflow expects on import, with no further conversion required.
- `clarifai_output`: the original ZIP downloaded from Clarifai. You can ignore this now, it was just the intermediate download before extraction.
With the conversion complete, the next step is importing the dataset into Roboflow and verifying that all annotations transferred correctly.
Import COCO JSON Dataset into Roboflow
With your COCO dataset ready in `coco_output/`, you can now import it into Roboflow. The process takes a few clicks and no additional conversion.
1. Create a New Project
Log in to Roboflow and create a new project. Select your annotation type (Object Detection for bounding boxes) and give your project a name.
2. Upload the Dataset
From your new project, click Upload Data. Click Select Folder and choose the `coco_output/` folder. Roboflow will automatically detect the COCO format and map the annotations to your images.
Your dataset is now fully migrated from Clarifai to Roboflow. Your images, classes, and bounding box annotations are preserved exactly as they were, ready for versioning, augmentation, and training.
Key Takeaways: Get Started with Clarifai Alternative Roboflow
Clarifai exports datasets in protobuf format rather than COCO, so you need `clarifai-datautils` to convert the export before Roboflow can read it. The migration itself preserves your images, classes, and bounding box annotations exactly as they were. Once your dataset lands in Roboflow, you get dataset versioning, model-assisted labeling, RF-DETR training, and Roboflow Workflows without any additional setup.
Further reading
Below are a few related topics you might be interested in:
Cite this Post
Use the following entry to cite this post in your research:
Mostafa Ibrahim. (May 20, 2026). Clarifai Alternative: How to Migrate Your Dataset to Roboflow. Roboflow Blog: https://blog.roboflow.com/migrate-clarifai-to-roboflow/