[?] How to export empty YOLOv5 labels files for images with no annotations

[hassanbadawy]

Describe the problem

when I run
split_view.export(
export_dir=export_dir,
dataset_type=fo.types.YOLOv5Dataset,
label_field=label_field,
split=split,
classes=classes,
)
it exports only all images that have annotation, but there are some images with no annotation don't export! I need my model to train on it to learn them as a background, without this the model gives me many false positive on those empty images.

System information

• OS Platform and Distribution (Linux Ubuntu 22.04):
• Python version (3.10.12):
• FiftyOne version (1.2.0):
• FiftyOne installed from (pip):

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?

• [x ] 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

[brimoor]

Hi @hassanbadawy 👋

Suppose you are exporting a field called ground_truth in YOLOv5 format. Then:

1. If ground_truth is None, then no labels TXT file is generated
2. If ground_truth is an empty detections object fo.Detections(), then an empty labels TXT file is generated

Sounds like you want option 2 in your case.

Here's an example:

import fiftyone as fo
import fiftyone.zoo as foz
import fiftyone.utils.random as four

dataset = foz.load_zoo_dataset("quickstart")

classes = dataset.distinct("ground_truth.detections.label")

# Create splits
dataset.untag_samples(dataset.distinct("tags"))
four.random_split(dataset, {"train": 0.7, "test": 0.3})

#
# Simulate a case where 50% of images have no annotations
#

# If `ground_truth` is `None`, no labels files are generated
# dataset.take(100).clear_sample_field("ground_truth")

# If `ground_truth` contains empty detections `fo.Detections()`, then empty
# labels files are generated
dataset.take(100).set_field("ground_truth", fo.Detections()).save()

# Export in YOLOv5 format
for split in ("train", "test"):
    split_view = dataset.match_tags(split)
    split_view.export(
        export_dir="/tmp/yolov5",
        dataset_type=fo.types.YOLOv5Dataset,
        label_field="ground_truth",
        split=split,
        classes=classes,
    )

find /tmp/yolov5/images/train -type f | wc -l
find /tmp/yolov5/labels/train -type f | wc -l
find /tmp/yolov5/images/test -type f | wc -l
find /tmp/yolov5/labels/test -type f | wc -l

140
140
60
60