Prediction with Segm Masks?

[cmbean2]

Hi,

I am working with a Detectron2 based model and I care about the segmentation mask predictions for my features. I have large images/regions with lots of small features and since my model was trained on small slices, SAHI seems to be the ideal way to improve/perform at all my predictions on my large images.

However, when I try to run the predict command it gives out no predictions on my images (neither bbox or segm). I don't get any errors thrown simply no predictions.

I am at a loss on how to troubleshoot it any help would be highly appreciated.
Below is the majority of my attempt to run SAHI, excluding the initialization of the model in the notebook I am using.

# will be used for detectron2 fasterrcnn model zoo name
from sahi.utils.detectron2 import Detectron2TestConstants

# import required functions, classes
from sahi import AutoDetectionModel
from sahi.predict import get_sliced_prediction, predict, get_prediction
from sahi.utils.file import download_from_url
from sahi.utils.cv import read_image
from IPython.display import Image
from sahi.models.detectron2 import Detectron2DetectionModel
from sahi.utils.cv import read_image
from sahi.utils.detectron2 import Detectron2TestConstants
from sahi.utils.import_utils import get_package_info
from sahi.models.base import DetectionModel

# create desired Detectron2 config
from detectron2.config import get_cfg
cfg = get_cfg()
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 1
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.3

# export config as yaml
from sahi.utils.detectron2 import export_cfg_as_yaml 
export_cfg_as_yaml(cfg, export_path='config.yaml')

model_path = "./model_manual_save.pth"
cfg_path = "./config.yaml"
print(model_path)
print(cfg_path)

detection_model = Detectron2DetectionModel(
    model_path = model_path,
    config_path = cfg_path,
    confidence_threshold=0.3,
    device="cuda:0",
)

After this I get an output that may be where my issue is but I can't say I know if this is right or wrong:

result = get_sliced_prediction(
    "./Test Regions/718DP_BlN_2_cropped_CVAT.tif",
    detection_model,
    slice_height = 256,
    slice_width = 256,
    overlap_height_ratio = 0.2,
    overlap_width_ratio = 0.2,
)

result.export_visuals(export_dir="./")

Image("./prediction_visual.png")

I get nothing out here at this point. Simply the original image loaded in and there are no annotations that are in the results either.