References that answer the question you actually searched.

BBBC010 dataset: C. elegans live/dead assay images with worm masks

BBBC010 from the Broad Bioimage Benchmark Collection: 100 brightfield and GFP images of C. elegans from an anti-infective screen, with foreground masks and one mask per worm. CC0.

Tomato leaf disease dataset for annotation: CCMT, 5,325 labeled field photos

An agriculture dataset you can browse: 5,325 tomato leaf photos from farms in Ghana, each labeled as healthy, leaf blight, leaf curl, septoria leaf spot or verticillium wilt. CC BY 4.0, from the CCMT crop pest and disease dataset.

COCO person keypoints dataset: 605 openly licensed images you can browse

A hosted subset of COCO 2017 val person keypoints with only openly licensed images: 605 images, 1,609 people with 17 keypoints, visibility flags and boxes, every image credited. Browse it in the browser or download it.

Keypoint annotation tools compared: CVAT, Label Studio, Roboflow and DataTorch

An honest comparison of four keypoint annotation tools: how each defines a skeleton, handles occluded points, exports COCO keypoints, and where each one is the better choice.

Keypoint formats compared: COCO 17, MPII 16, MediaPipe 33 and COCO-WholeBody 133

How the main human pose keypoint formats differ: how many points, which body parts, the index order of each, and Python code to convert between COCO, MPII, MediaPipe and COCO-WholeBody.

Multi-channel TIFF tutorial: per-channel lookup tables on 7-channel iFISH images

Open a seven-channel, 16-bit FISH microscopy TIFF in the browser, set a lookup table per channel, and check spot labels against the raw signal, using a hosted IDR dataset (CC BY 4.0).

BBBC041 dataset: P. vivax malaria blood smears with labeled cells

BBBC041 from the Broad Bioimage Benchmark Collection: 1,364 Giemsa-stained blood smear images with about 80,000 cells in bounding boxes, six classes from RBCs to schizonts. CC BY-NC-SA 3.0.

BBBC005 dataset: synthetic fluorescent cells with known counts and focus blur

BBBC005 from the Broad Bioimage Benchmark Collection: 19,200 synthetic 8-bit fluorescence images with 1 to 100 cells each, blur levels 1 to 48, and foreground masks for the in-focus images. CC0.

Run a Hugging Face model on your images from the annotator

Wire a Hugging Face model to a button in the DataTorch annotator. Press it on an image and the model's boxes or polygons land on the file, ready to correct.

Release 0.1.0: Pipelines, rebuilt

Pipelines gets a visual editor and an annotator button that runs your model and puts its results on the file. The agent daemon is now a runner.

Nuclei segmentation tutorial with the DSB2018 dataset

Browse the DSB2018 nuclei dataset in your browser, outline nuclei on your own microscopy images, and export COCO polygons you can turn into instance masks for training.

What is keypoint annotation?

What keypoints, skeletons and visibility flags are, how they differ from boxes and masks, then the workflow: schema, placing points, occlusion and left/right, review, export.

How to annotate keypoints in DataTorch

Step by step: put the keypoint schema and skeleton on a label, place points in order with the Keypoints tool, mark occluded points, edit with the Move tool, export to COCO.

The COCO keypoint format

All 17 keypoints in order, 0-based vs 1-based indexing, the official skeleton pairs, visibility flags, and worked annotation JSON.

COCO keypoint visualizer

Drag the 17 COCO person keypoints on a live skeleton, set visibility flags, paste an annotation to validate and render it, and copy valid COCO JSON out. Runs in the browser.

Keypoint and pose estimation datasets

Downloadable keypoint datasets hosted on DataTorch, plus the standard benchmarks (COCO, MPII, CrowdPose, AP-10K and more) with sizes and keypoint counts.

The DSB2018 nuclei segmentation dataset

What is inside the standard nuclei benchmark, the per-nucleus mask format, common pitfalls, and which model to train: U-Net, StarDist, Cellpose, or Mask R-CNN.