[BUG] issue viewing the bounding boxes from "rtdetr-l.pt" results

[LayanCS]

Hi,

I'm trying to run an inference using RT-DETR and FlagSeg dataset. I'm using FlagSeg dataset as is, and I tested it using both "yolov8s.pt" and "rtdetr-l.pt". The bounding boxes when using RT-DETR are squished in the upper left corner.

I cannot tell if this is an issue when converting the prediction boxes, or if I need to change the dataset type for RT-DETR.

This is how I'm loading my dataset:

dataset_image_dir = "/FlagSeg/test_images" # dataset directory
dataset_labels = "/FlagSeg/test_dataset.json"

dataset = fo.Dataset.from_dir(
    dataset_type=fo.types.COCODetectionDataset,
    data_path=dataset_image_dir,
    labels_path=dataset_labels,
    name="FlagSeg",
    max_samples=20,
    shuffle=True,
    seed=51
)

Model:

model = YOLO("rtdetr-l.pt") # or yolov8s.pt
dataset.apply_model(model, label_field="boxes")

Result in the app: (Looking at the json file of the image I can see the prediction boxes with very small values)

Zooming in:

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 22.04): MacOS - Sequoia
• Python version (python --version): Python 3.9.21
• FiftyOne version (fiftyone --version): FiftyOne v1.3.0, Voxel51, Inc.
• FiftyOne installed from (pip or source): Pip
• Ultralytics Version: 8.3.75

Willingness to contribute

The FiftyOne Community encourages bug fix contributions. Would you or another
member of your organization be willing to contribute a fix for this bug to the
FiftyOne codebase?

• Yes. I can contribute a fix for this bug independently
• Yes. I would be willing to contribute a fix for this bug with guidance
  from the FiftyOne community
• No. I cannot contribute a bug fix at this time