Guide

How to use MLTools

Every tool follows the same layout: controls on the left, the work on the right, an export panel at the bottom of the controls. Files are read by the page and never uploaded; annotation projects and drafts are saved in this browser's storage (IndexedDB), keyed by file name and size, so re-dropping the same folder restores your labels. Clearing site data removes them, so export when a session ends.

Four common jobs, start to finish

Build a YOLO detection dataset from a folder of photos

Open the Bounding Box Annotator, drop the folder, type your classes and draw boxes (number keys switch class, arrows move between images). Labels save in the browser as you go.
Export YOLO: you get labels/<image>.txt and data.yaml. Zip them with your images.
Drop that zip into the Dataset Validator to catch class-id and bounds problems, then into the Duplicate Image Finder to remove copies.
Drop it into the Train / Validation / Test Splitter (80/10/10, seed 42, stratified). The output has train/ val/ test/ and a data.yaml.
Open the Fine-tuning Script Generator, choose YOLO, paste the class names, download train.py, and run it on your GPU machine.
After training, export predictions at a low confidence and compare with the ground truth in the mAP Calculator.

Bounding Box Annotator

Prepare a PaddleOCR or recognition dataset

Open the OCR Text Annotator and drop the images. Drag a box around each text line; drag corners for skewed text.
Type the transcription, or press Suggest to run Tesseract on the crop in your browser and correct it. Tab moves to the next box.
Export PaddleOCR Label.txt for detection training, or Recognition crops (perspective-corrected PNGs plus rec_gt.txt) for a recognition model.
Split with the Dataset Splitter and check class balance and text lengths in Dataset Statistics.

OCR Text Annotator

Plan and run a LoRA fine-tune of a 7–8B model

Assemble examples in the LLM Fine-tuning Dataset Builder (type them, or import CSV or JSONL). Every row is validated; export OpenAI-format JSONL with a train/val split.
Count tokens in the Tokenizer Playground, then open the GPU Memory Calculator: pick the model preset, LoRA or QLoRA, batch and sequence length, and read which cards it fits.
Estimate hours and cost in the Training Time & Cost Estimator with a measured or preset throughput.
Generate train.py in the Fine-tuning Script Generator (LLM LoRA recipe), download the zip and run it where the GPU is.
Check the exact prompt string the model will see with the Chat Template Converter before inference.

LLM Dataset Builder

Understand an ONNX export before deploying it

Drop the .onnx file into the ONNX Model Inspector: input and output names, shapes and dynamic axes, operator counts, largest weights, opset.
For YOLO exports, read the class names, stride and imgsz from the metadata section to set up post-processing.
Estimate memory for a quantised deployment in the Model Size & Quantization Calculator.

ONNX Model Inspector

Formats each tool reads and writes

Tool	Reads	Writes
Bounding Box Annotator	images; YOLO txt + data.yaml; COCO json; VOC xml; CSV	YOLO zip, COCO json, VOC zip, CSV
Polygon Annotator	images; COCO json; YOLO-seg txt	COCO json, YOLO-seg zip
OCR Text Annotator	images; PaddleOCR Label.txt	Label.txt, ICDAR gt zip, recognition crops zip
Image / Text Classification Labelers	images + labels.csv; CSV / JSONL / lines	CSV, JSONL, class-folder zip, label map
NER Annotator	text, JSONL (entities / label)	spaCy JSONL, Doccano JSONL, BIO, CoNLL
Splitter / Converter / Stats / Validator	YOLO, COCO, VOC, CSV boxes, class folders, CSV / JSONL tables, zips	split zips with data.yaml, converted labels, reports, cleaned datasets
LLM Dataset Builder	CSV; OpenAI, ShareGPT, Alpaca, prompt/completion JSONL	OpenAI, Alpaca, ShareGPT JSONL
Script Generator	your settings	train.py, data.yaml / config, requirements.txt, run.sh
ONNX Inspector	.onnx	JSON / Markdown summary
Developer tools	JSON, YAML, TOML, CSV, JSONL, .env, Markdown, images	the same, converted or formatted

Keyboard shortcuts in the annotators

Boxes and polygons

1–9 class · ← → image · Del remove selected · Z undo · Esc deselect · Ctrl+wheel zoom · wheel pan · Space+drag pan · polygons: Enter close, Backspace last point, Shift straight edges, Alt+click insert or delete a vertex.

Classification and OCR

Classification: 1–9 label and advance, Space skip, Backspace back. OCR: Tab next box, Esc leave the text field, Del remove the selected box.

Every tool

Annotation

Bounding Box Annotator — Draw, move and resize boxes on a folder of images and export YOLO, COCO, Pascal VOC or CSV labels.
Polygon & Segmentation Annotator — Trace polygons for instance segmentation and export COCO segmentation or YOLO-seg labels.
OCR Text Annotator — Draw text boxes, type or auto-suggest the transcription, and export PaddleOCR, ICDAR or recognition-crop datasets.
Image Classification Labeler — Assign one class per image with the keyboard, then export a CSV or a class-folder zip.
Text Classification Labeler — Label rows of text with keyboard shortcuts and export CSV or JSONL for a classifier.
Entity (NER) Span Annotator — Select spans in text, tag them with entity types, and export spaCy or BIO token labels.

Datasets

Train / Validation / Test Splitter — Split images with labels, class folders, or CSV/JSONL rows into train, val and test with a fixed seed and stratification.
Label Format Converter — Convert detection labels between YOLO, COCO, Pascal VOC and CSV without losing a box.
Dataset Statistics — Class counts, box sizes, aspect ratios and image sizes for a labelled dataset, drawn as charts.
Dataset Validator — Find missing labels, empty files, bad class ids, out-of-bounds boxes and unreadable images before a training run fails.
Duplicate Image Finder — Find exact and near-duplicate images with perceptual hashing and remove them from a dataset.
Batch Resize & Augment — Resize, letterbox, flip, rotate and adjust a folder of images in the browser, with YOLO labels transformed to match.
LLM Fine-tuning Dataset Builder — Assemble chat and instruction examples, validate them, count tokens and export OpenAI, Alpaca or ShareGPT JSONL.
JSONL Viewer & Validator — Open large JSONL files, see the schema and statistics, find bad lines and duplicates, filter and convert to CSV.

Fine-tuning

GPU Memory (VRAM) Calculator — Estimate the GPU memory a fine-tune or inference run needs from parameters, precision, LoRA, batch size and sequence length.
Fine-tuning Script Generator — Generate a runnable training script and config for YOLO detection, Hugging Face LoRA SFT, or an image classifier from your paths and settings.
Training Time & Cost Estimator — Estimate hours and cloud cost from dataset size, epochs, throughput and GPU price, for LLM and vision runs.
Learning Rate Schedule Visualizer — Plot cosine, linear, step and one-cycle schedules with warmup over your step count and export the values.
Model Size & Quantization Calculator — File size and inference memory for a model at fp32, fp16, int8, int4 and GGUF quantisation levels, plus KV cache per token.
Classification Metrics & Confusion Matrix — Paste true and predicted labels and get accuracy, precision, recall, F1 per class, macro and weighted averages, and a confusion matrix.
mAP Calculator for Object Detection — Compute AP per class and mAP@0.5 and mAP@0.5:0.95 from ground-truth and prediction files in YOLO or COCO format.

Models

ONNX Model Inspector — Open an ONNX file and see inputs, outputs, shapes, operator counts, initializer sizes and metadata, without installing anything.
Tokenizer Playground — See how text splits into tokens for GPT-style tokenizers, count tokens for a file, and estimate API cost.
Chat Template Converter — Turn a system/user/assistant conversation into the exact prompt string a model was trained on: ChatML, Llama 3, Mistral, Gemma, Alpaca or Vicuna.

Developer

JSON Formatter, Validator & Tree Viewer — Format, validate, minify and explore JSON as a collapsible tree, with a path query and error positions.
YAML, JSON & TOML Converter — Convert between YAML, JSON and TOML with validation, and see exactly where a config file breaks.
CSV, JSON & JSONL Converter — Convert CSV to JSON or JSONL and back, with delimiter detection, type inference, column statistics and a preview table.
JSON Schema Generator & Validator — Generate a JSON Schema from sample data and validate documents against a schema with precise error paths.
Regex Tester — Test regular expressions against text with live highlighting, group capture, replacement preview and a flag guide.
Diff Checker — Compare two texts, JSON documents or config files line by line or word by word, side by side or inline.
Encoders, Hashes & IDs — Base64, URL and hex encoding, JWT decoding, MD5, SHA-1, SHA-256 and SHA-512 hashes, and UUID or ULID generation.
Cron Expression Builder — Build or explain a cron expression, see the next run times in your timezone and in UTC, and copy crontab or Kubernetes lines.
Timestamp & Timezone Converter — Convert Unix seconds and milliseconds, ISO 8601 and human dates between UTC, IST and any timezone, and compute durations.
.env, JSON & Compose Environment Converter — Convert between .env files, JSON objects, docker-compose environment lists, Kubernetes ConfigMaps and shell exports.
Markdown Editor & Preview — Write Markdown with a live preview, tables and code highlighting, and export clean HTML or a model card.
Image Inspector, EXIF & Base64 — Read an image's dimensions, format, EXIF and orientation, strip metadata, and convert to and from Base64 data URLs.

36 tools. Missing one you need? Email ahzushan@gmail.com.