EVF-SAM2 Segmentation

EVF-SAM2 Segmentation delivers precise, prompt-driven image segmentation for any object.

A tradução deste card ainda não está disponível; parte do conteúdo aparece em inglês.

Sobre este modelo

EVF-SAM2 Segmentation is a state-of-the-art visual foundation model designed for interactive instance and panoptic segmentation. It accepts text prompts, bounding boxes, or point clicks to generate high-quality segmentation masks across diverse categories. The model is particularly effective in zero-shot scenarios, requiring no fine-tuning for new classes.

Use EVF-SAM2 when you need to extract specific objects from photos, create masks for editing, or annotate datasets quickly. Its distinctive strength lies in combining language-guided understanding with pixel-level precision, enabling intuitive control even for complex scenes with overlapping or partially visible targets.

Melhores casos de uso

Remove backgrounds from product photos
Isolate subjects for portrait editing
Create masks for object removal in scenes
Annotate training data for computer vision
Generate cutouts for collage or compositing

Pontos fortes

Prompt flexibility

Supports text, points, and boxes for intuitive segmentation.

High accuracy

Produces clean, detailed masks even on challenging edges.

Zero-shot capability

Segments arbitrary objects without class-specific training.

Dicas

Use specific nouns in text prompts for better targeting
Provide multiple positive points for complex shapes
Avoid vague descriptions like 'the thing' or 'stuff'
Zoom in on edges for refined mask adjustments

Limitações

Struggles with very fine hair or transparent objects
May misinterpret ambiguous prompts in cluttered scenes
Requires sufficient visual contrast for optimal results

Exemplos de geração

Exemplo em breve

Portrait background removal

Separate the subject from the scenic background with clean edges

Exemplo em breve

Product cutout for e-commerce

Generate a precise mask for the sneaker on a white backdrop

Exemplo em breve

Complex scene annotation

Identify and mask all vehicles in the busy intersection

Prompts de exemplo

Isolate a dog in a park

Segment the golden retriever playing fetch

Extract a car from street view

Create a mask for the red sedan parked on the curb

Select a person in a crowd

Segment the woman wearing a blue jacket in the center

Parâmetros

prompt

Text description or visual cue used to guide which object to segment

box_coords

Bounding box defining the region of interest

point_coords

Coordinates of clicks indicating object location

multimask_output

Whether to return multiple candidate masks for ambiguous prompts

Preços

$0.0200 por requisição

Os preços exibidos são o que você paga na Muitto. Sem surpresas.

Perguntas frequentes

Can EVF-SAM2 segment multiple objects at once?

Yes, it supports panoptic segmentation to label and mask all instances in a scene simultaneously.

Do I need to train the model for my specific use case?

No, EVF-SAM2 works out of the box with zero-shot prompting for most common objects.

What input formats are supported for prompts?

You can use natural language text, bounding boxes, or interactive point clicks to guide segmentation.