[MohammadHijjawi97]

RecognitionDataset.merge_dataset moves the root of the merged dataset to / and joins each image path to its former root. With a relative root (e.g. --train_path data/train in references/recognition/train.py, which merges every subfolder), the joined paths stay relative and are then read from the filesystem root, so every sample fails to load.

Repro:

# cwd contains train/part_a/{images,labels.json} and train/part_b/{images,labels.json}
ds = RecognitionDataset("train/part_a/images", "train/part_a/labels.json")
ds.merge_dataset(RecognitionDataset("train/part_b/images", "train/part_b/labels.json"))
ds[0]

Before: FileNotFoundError for /train/part_a/images/part_a.png (resolved from the filesystem root instead of the working directory)
After: both samples load.

Changes:

• make both roots absolute before joining them to the image paths
• add test_recognition_dataset_merge (relative image folder, two merges, every sample readable); it fails without the fix

Checks: pytest tests/pytorch/test_datasets_pt.py -k recognition_dataset, ruff check, ruff format --check, mypy.

Any feedback is welcome 🤗

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.63%. Comparing base (177c3c0) to head (77d79e5).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2177   +/-   ##
=======================================
  Coverage   97.62%   97.63%           
=======================================
  Files         170      170           
  Lines       10744    10744           
=======================================
+ Hits        10489    10490    +1     
+ Misses        255      254    -1     

Flag	Coverage Δ
unittests	97.63% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[felixdittrich92]

Thanks for the PR @MohammadHijjawi97 👍

Only one small change and we are good to merge.

[MohammadHijjawi97]

Thanks for the review, @felixdittrich92! I removed the separate test and extended test_recognition_dataset instead. It now builds the dataset from a relative image folder, merges two more datasets into it, and iterates a DataLoader over the merged dataset, checking the batch shapes and that all labels come through.

[felixdittrich92]

Thanks @MohammadHijjawi97 LGTM 👍