Implementation:Obss Sahi Get Prediction Full Image

Knowledge Sources	SAHI SAHI Docs
Domains	Object_Detection, Computer_Vision, Inference
Last Updated	2026-02-08 12:00 GMT

Overview

Concrete tool for running full-image detection as a complement to sliced detection, reusing the get_prediction() function with zero shift offset, provided by the SAHI library.

Description

Full-image detection in SAHI uses the same get_prediction() function as per-slice detection, but with shift_amount=[0, 0] and full_shape=None (inferred from the image). This pass is invoked within get_sliced_prediction() at sahi/predict.py:L313-326 when perform_standard_pred=True and the image was sliced into multiple tiles.

The full-image result is appended to the per-slice results and the combined set is passed to the postprocessing (merging) step. This ensures large objects are detected at their natural scale.

Usage

This implementation is invoked automatically within the sliced prediction pipeline. It uses the same get_prediction() API with different parameters. Use it explicitly when you want to combine full-image and slice-based detections manually.

Code Reference

Source Location

• Repository: sahi
• File: sahi/predict.py
• Lines: L56-135 (same function as per-slice detection)
• Invocation context: L313-326 within get_sliced_prediction()

Signature

def get_prediction(
    image,
    detection_model,
    shift_amount: list | None = None,  # [0, 0] for full image
    full_shape=None,                    # inferred from image
    postprocess: PostprocessPredictions | None = None,
    verbose: int = 0,
    exclude_classes_by_name: list[str] | None = None,
    exclude_classes_by_id: list[int] | None = None,
) -> PredictionResult:
    """Perform prediction on the full image (no slicing).

    For full-image detection, shift_amount defaults to [0, 0]
    and full_shape is inferred from the input image dimensions.
    """

Import

from sahi.predict import get_prediction

I/O Contract

Inputs

Name	Type	Required	Description
image	str or np.ndarray	Yes	Full original image (path or numpy array)
detection_model	DetectionModel	Yes	Same initialized model used for per-slice detection
shift_amount	list[int]	No	Always [0, 0] for full-image detection (no remapping)
full_shape	list[int]	No	None (inferred from image dimensions)

Outputs

Name	Type	Description
return	PredictionResult	Contains .object_prediction_list with full-image detections (coordinates already in full-image space)

Usage Examples

Full-Image Detection Pass

from sahi import AutoDetectionModel
from sahi.predict import get_prediction

model = AutoDetectionModel.from_pretrained(
    model_type="ultralytics",
    model_path="yolov8n.pt",
    device="cuda:0",
)

# Full-image detection (no shift needed)
full_result = get_prediction(
    image="path/to/large_image.jpg",
    detection_model=model,
    shift_amount=[0, 0],
)

print(f"Full-image detections: {len(full_result.object_prediction_list)}")

Combining with Slice Detections

from sahi.predict import get_prediction
from sahi.slicing import slice_image

# Slice the image
slice_result = slice_image(image="large.jpg", slice_height=640, slice_width=640)

# Collect per-slice predictions
all_preds = []
for i, img in enumerate(slice_result.images):
    result = get_prediction(
        image=img,
        detection_model=model,
        shift_amount=slice_result.starting_pixels[i],
        full_shape=[slice_result.original_image_height, slice_result.original_image_width],
    )
    all_preds.extend(result.object_prediction_list)

# Add full-image detection
full_result = get_prediction(
    image="large.jpg",
    detection_model=model,
    shift_amount=[0, 0],
)
all_preds.extend(full_result.object_prediction_list)

# all_preds now ready for merging step

Related Pages

Implements Principle

• Principle:Obss_Sahi_Full_Image_Detection

Requires Environment

• Environment:Obss_Sahi_Python_Detection_Frameworks