huggingface-vision-trainer

Vision Model Training on Hugging Face Jobs

Train object detection, image classification, and SAM/SAM2 segmentation models on managed cloud GPUs. No local GPU setup required—results are automatically saved to the Hugging Face Hub.

When to Use This Skill

Use this skill when users want to:

• Fine-tune object detection models (D-FINE, RT-DETR v2, DETR, YOLOS) on cloud GPUs or local
• Fine-tune image classification models (timm: MobileNetV3, MobileViT, ResNet, ViT/DINOv3, or any Transformers classifier) on cloud GPUs or local
• Fine-tune SAM or SAM2 models for segmentation / image matting using bbox or point prompts
• Train bounding-box detectors on custom datasets
• Train image classifiers on custom datasets
• Train segmentation models on custom mask datasets with prompts
• Run vision training jobs on Hugging Face Jobs infrastructure
• Ensure trained vision models are permanently saved to the Hub

Related Skills

• hugging-face-jobs — General HF Jobs infrastructure: token authentication, hardware flavors, timeout management, cost estimation, secrets, environment variables, scheduled jobs, and result persistence. Refer to the Jobs skill for any non-training-specific Jobs questions (e.g., "how do secrets work?", "what hardware is available?", "how do I pass tokens?").
• hugging-face-model-trainer — TRL-based language model training (SFT, DPO, GRPO). Use that skill for text/language model fine-tuning.

Local Script Execution

Helper scripts use PEP 723 inline dependencies. Run them with uv run:

uv run scripts/dataset_inspector.py --dataset username/dataset-name --split train
uv run scripts/estimate_cost.py --help

Prerequisites Checklist

Before starting any training job, verify:

Account & Authentication

• Hugging Face Account with Pro, Team, or Enterprise plan (Jobs require paid plan)
• Authenticated login: Check with hf_whoami() (tool) or hf auth whoami (terminal)
• Token has write permissions
• MUST pass token in job secrets — see directive #3 below for syntax (MCP tool vs Python API)

Dataset Requirements — Object Detection

• Dataset must exist on Hub
• Annotations must use the objects column with bbox, category (and optionally area) sub-fields
• Bboxes can be in xywh (COCO) or xyxy (Pascal VOC) format — auto-detected and converted
• Categories can be integers or strings — strings are auto-remapped to integer IDs
• image_id column is optional — generated automatically if missing
• ALWAYS validate unknown datasets before GPU training (see Dataset Validation section)

Dataset Requirements — Image Classification

• Dataset must exist on Hub
• Must have an image column (PIL images) and a label column (integer class IDs or strings)
• The label column can be ClassLabel type (with names) or plain integers/strings — strings are auto-remapped
• Common column names auto-detected: label, labels, class, fine_label
• ALWAYS validate unknown datasets before GPU training (see Dataset Validation section)

Dataset Requirements — SAM/SAM2 Segmentation

• Dataset must exist on Hub
• Must have an image column (PIL images) and a mask column (binary ground-truth segmentation mask)
• Must have a prompt — either:
  • A prompt column with JSON containing {"bbox": [x0,y0,x1,y1]} or {"point": [x,y]}
  • OR a dedicated bbox column with [x0,y0,x1,y1] values
  • OR a dedicated point column with [x,y] or [[x,y],...] values
• Bboxes should be in xyxy format (absolute pixel coordinates)
• Example dataset: merve/MicroMat-mini (image matting with bbox prompts)
• ALWAYS validate unknown datasets before GPU training (see Dataset Validation section)

Critical Settings

• Timeout must exceed expected training time — Default 30min is TOO SHORT. See directive #6 for recommended values.
• Hub push must be enabled — push_to_hub=True, hub_model_id="username/model-name", token in secrets

Dataset Validation

Validate dataset format BEFORE launching GPU training to prevent the #1 cause of training failures: format mismatches.

ALWAYS validate for unknown/custom datasets or any dataset you haven't trained with before. Skip for cppe-5 (the default in the training script).

Running the Inspector

Option 1: Via HF Jobs (recommended — avoids local SSL/dependency issues):

hf_jobs("uv"

Option 2: Locally:

Option 3: Via HfApi().run_uv_job() (if hf_jobs MCP unavailable):

Reading Results

• ✓ READY — Dataset is compatible, use directly
• ✗ NEEDS FORMATTING — Needs preprocessing (mapping code provided in output)

Automatic Bbox Preprocessing

The object detection training script (scripts/object_detection_training.py) automatically handles bbox format detection (xyxy→xywh conversion), bbox sanitization, image_id generation, string category→integer remapping, and dataset truncation. No manual preprocessing needed — just ensure the dataset has objects.bbox and objects.category columns.

Training workflow

Copy this checklist and track progress:

Step 1: Verify prerequisites

Follow the Prerequisites Checklist above.

Step 2: Validate dataset

Run the dataset inspector BEFORE spending GPU time. See "Dataset Validation" section above.

Step 3: Ask user preferences

ALWAYS use the AskUserQuestion tool with option-style format:

Step 4: Prepare training script

For object detection, use scripts/object_detection_training.py as the production-ready template. For image classification, use scripts/image_classification_training.py. For SAM/SAM2 segmentation, use scripts/sam_segmentation_training.py. All scripts use HfArgumentParser — all configuration is passed via CLI arguments in script_args, NOT by editing Python variables. For timm model details, see references/timm_trainer.md. For SAM2 training details, see references/finetune_sam2_trainer.md.

Step 5: Save script, submit job, and report

1. Save the script locally to submitted_jobs/ in the workspace root (create if needed) with a descriptive name like training_<dataset>_<YYYYMMDD_HHMMSS>.py. Tell the user the path.
2. Submit using hf_jobs MCP tool (preferred) or HfApi().run_uv_job() — see directive #1 for both methods. Pass all config via script_args.
3. Report the job ID (from .id attribute), monitoring URL, Trackio dashboard (https://huggingface.co/spaces/{username}/trackio), expected time, and estimated cost.
4. Wait for user to request status checks — don't poll automatically. Training jobs run asynchronously and can take hours.

Critical directives

These rules prevent common failures. Follow them exactly.

1. Job submission: hf_jobs MCP tool vs Python API

hf_jobs() is an MCP tool, NOT a Python function. Do NOT try to import it from huggingface_hub. Call it as a tool:

If hf_jobs MCP tool is unavailable, use the Python API directly:

Critical differences between the two methods:

Rules for both methods:

• The training script MUST include PEP 723 inline metadata with dependencies
• Do NOT use image or command parameters (those belong to run_job(), not run_uv_job())

2. Authentication via job secrets + explicit hub_token injection

Job config MUST include the token in secrets — syntax depends on submission method (see table above).

Training script requirement: The Transformers Trainer calls create_repo(token=self.args.hub_token) during __init__() when push_to_hub=True. The training script MUST inject HF_TOKEN into training_args.hub_token AFTER parsing args but BEFORE creating the Trainer. The template scripts/object_detection_training.py already includes this:

If you write a custom script, you MUST include this token injection before the Trainer(...) call.

• Do NOT call login() in custom scripts unless replicating the full pattern from scripts/object_detection_training.py
• Do NOT rely on implicit token resolution (hub_token=None) — unreliable in Jobs
• See the hugging-face-jobs skill → Token Usage Guide for full details

3. JobInfo attribute

Access the job identifier using .id (NOT .job_id or .name — these don't exist):

4. Required training flags and HfArgumentParser boolean syntax

scripts/object_detection_training.py uses HfArgumentParser — all config is passed via script_args. Boolean arguments have two syntaxes:

• bool fields (e.g., push_to_hub, do_train): Use as bare flags (--push_to_hub) or negate with --no_ prefix (--no_remove_unused_columns)
• Optional[bool] fields (e.g., greater_is_better): MUST pass explicit value (--greater_is_better True). Bare --greater_is_better causes error: expected one argument

Required flags for object detection:

Required flags for image classification:

Required flags for SAM/SAM2 segmentation:

5. Timeout management

Default 30 min is TOO SHORT for object detection. Set minimum 2-4 hours. Add 30% buffer for model loading, preprocessing, and Hub push.

6. Trackio monitoring

Trackio is always enabled in the object detection training script — it calls trackio.init() and trackio.finish() automatically. No need to pass --report_to trackio. The project name is taken from --output_dir and the run name from --run_name. For image classification, pass --report_to trackio in TrainingArguments.

Dashboard at: https://huggingface.co/spaces/{username}/trackio

Model & hardware selection

Recommended object detection models

Start with ustc-community/dfine-small-coco for fast iteration. Move to D-FINE Large or RT-DETR v2 R50 for better accuracy.

Recommended image classification models

All timm/ models work out of the box via AutoModelForImageClassification (loaded as TimmWrapperForImageClassification). See references/timm_trainer.md for details.

Start with timm/mobilenetv3_small_100.lamb_in1k for fast iteration. Move to timm/resnet50.a1_in1k or timm/vit_base_patch16_dinov3.lvd1689m for better accuracy.

Recommended SAM/SAM2 segmentation models

Start with facebook/sam2.1-hiera-small for fast iteration. SAM2 models are generally more efficient than SAM v1 at similar quality. Only the mask decoder is trained by default (vision and prompt encoders are frozen).

Hardware recommendation

All recommended OD and IC models are under 100M params — t4-small (16 GB VRAM, $0.40/hr) is sufficient for all of them. Image classification models are generally smaller and faster than object detection models — t4-small handles even ViT-Base comfortably. For SAM2 models up to hiera-base-plus, t4-small is sufficient since only the mask decoder is trained. For sam2.1-hiera-large or SAM v1 models, use l4x1 or a10g-large. Only upgrade if you hit OOM from large batch sizes — reduce batch size first before switching hardware. Common upgrade path: t4-small → l4x1 ($0.80/hr, 24 GB) → a10g-large ($1.50/hr, 24 GB).

For full hardware flavor list: refer to the hugging-face-jobs skill. For cost estimation: run scripts/estimate_cost.py.

Quick start — Object Detection

The script_args below are the same for both submission methods. See directive #1 for the critical differences between them.

Key OD script_args

• --model_name_or_path — recommended: "ustc-community/dfine-small-coco" (see model table above)
• --dataset_name — the Hub dataset ID
• --image_square_size — 480 (fast iteration) or 800 (better accuracy)
• --hub_model_id — "username/model-name" for Hub persistence
• --num_train_epochs — 30 typical for convergence
• --train_val_split — fraction to split for validation (default 0.15), set if dataset lacks a validation split
• --max_train_samples — truncate training set (useful for quick test runs, e.g. "785" for ~10% of a 7.8K dataset)
• --max_eval_samples — truncate evaluation set

Quick start — Image Classification

Key IC script_args

• --model_name_or_path — any timm/ model or Transformers classification model (see model table above)
• --dataset_name — the Hub dataset ID
• --image_column_name — column containing PIL images (default: "image")
• --label_column_name — column containing class labels (default: "label")
• --hub_model_id — "username/model-name" for Hub persistence
• --num_train_epochs — 3-5 typical for classification (fewer than OD)
• --per_device_train_batch_size — 16-64 (classification models use less memory than OD)
• --train_val_split — fraction to split for validation (default 0.15), set if dataset lacks a validation split
• --max_train_samples / --max_eval_samples — truncate for quick tests

Quick start — SAM/SAM2 Segmentation

Key SAM script_args

• --model_name_or_path — SAM or SAM2 model (see model table above); auto-detects SAM vs SAM2
• --dataset_name — the Hub dataset ID (e.g., "merve/MicroMat-mini")
• --prompt_type — "bbox" or "point" — type of prompt in the dataset
• --prompt_column_name — column with JSON-encoded prompts (default: "prompt")
• --bbox_column_name — dedicated bbox column (alternative to JSON prompt column)
• --point_column_name — dedicated point column (alternative to JSON prompt column)
• --mask_column_name — column with ground-truth masks (default: "mask")
• --hub_model_id — "username/model-name" for Hub persistence
• --num_train_epochs — 20-30 typical for SAM fine-tuning
• --per_device_train_batch_size — 2-4 (SAM models use significant memory)
• --freeze_vision_encoder / --freeze_prompt_encoder — freeze encoder weights (default: both frozen, only mask decoder trains)
• --train_val_split — fraction to split for validation (default 0.1)

Checking job status

MCP tool (if available):

Python API fallback:

Common failure modes

OOM (CUDA out of memory)

Reduce per_device_train_batch_size (try 4, then 2), reduce IMAGE_SIZE, or upgrade hardware.

Dataset format errors

Run scripts/dataset_inspector.py first. The training script auto-detects xyxy vs xywh, converts string categories to integer IDs, and adds image_id if missing. Ensure objects.bbox contains 4-value coordinate lists in absolute pixels and objects.category contains either integer IDs or string labels.

Hub push failures (401)

Verify: (1) job secrets include token (see directive #2), (2) script sets training_args.hub_token BEFORE creating the Trainer, (3) push_to_hub=True is set, (4) correct hub_model_id, (5) token has write permissions.

Job timeout

Increase timeout (see directive #5 table), reduce epochs/dataset, or use checkpoint strategy with hub_strategy="every_save".

KeyError: 'test' (missing test split)

The object detection training script handles this gracefully — it falls back to the validation split. Ensure you're using the latest scripts/object_detection_training.py.

Single-class dataset: "iteration over a 0-d tensor"

torchmetrics.MeanAveragePrecision returns scalar (0-d) tensors for per-class metrics when there's only one class. The template scripts/object_detection_training.py handles this by calling .unsqueeze(0) on these tensors. Ensure you're using the latest template.

Poor detection performance (mAP < 0.15)

Increase epochs (30-50), ensure 500+ images, check per-class mAP for imbalanced classes, try different learning rates (1e-5 to 1e-4), increase image size.

For comprehensive troubleshooting: see references/reliability_principles.md

Reference files

• scripts/object_detection_training.py — Production-ready object detection training script
• scripts/image_classification_training.py — Production-ready image classification training script (supports timm models)
• scripts/sam_segmentation_training.py — Production-ready SAM/SAM2 segmentation training script (bbox & point prompts)
• scripts/dataset_inspector.py — Validate dataset format for OD, classification, and SAM segmentation
• scripts/estimate_cost.py — Estimate training costs for any vision model (includes SAM/SAM2)
• references/object_detection_training_notebook.md — Object detection training workflow, augmentation strategies, and training patterns
• references/image_classification_training_notebook.md — Image classification training workflow with ViT, preprocessing, and evaluation
• references/finetune_sam2_trainer.md — SAM2 fine-tuning walkthrough with MicroMat dataset, DiceCE loss, and Trainer integration
• references/timm_trainer.md — Using timm models with HF Trainer (TimmWrapper, transforms, full example)
• references/hub_saving.md — Detailed Hub persistence guide and verification checklist
• references/reliability_principles.md — Failure prevention principles from production experience

External links

• Transformers Object Detection Guide
• Transformers Image Classification Guide
• DETR Model Documentation
• ViT Model Documentation
• HF Jobs Guide — Main Jobs documentation
• HF Jobs Configuration — Hardware, secrets, timeouts, namespaces
• HF Jobs CLI Reference — Command line interface
• Object Detection Models
• Image Classification Models
• SAM2 Model Documentation
• SAM Model Documentation
• Object Detection Datasets
• Image Classification Datasets