CLI pm_evaluate detection crashes with AttributeError when --out_fname is used

[Harshavardhan-Raju]

File: perceptionmetrics/cli/evaluate.py
Line: 197

Description

TorchImageDetectionModel.eval() returns a dict with keys metrics_df
and metrics_factory. The CLI calls .to_csv() directly on this dict,
which raises AttributeError every time detection evaluation is run with
--out_fname.

Bug

results = model.eval(...)
results.to_csv(out_fname)  # results is a dict for detection models

Fix

if isinstance(results, dict):
    results = results["metrics_df"]
results.to_csv(out_fname)

Steps to Reproduce

1. Set up a COCO dataset and a torchscript detection model
2. Run evaluation with --out_fname:
   pm_evaluate detection image --model_format torch --model model.pt
   --dataset_format coco --dataset_dir /path/to/coco --out_fname results.csv
3. Evaluation completes but crashes before saving with:
   AttributeError: 'dict' object has no attribute 'to_csv'

[samscienced32-blip]

Need any help here? Otherwise, if you have fixed it, please close the issue, thanks.

[prawnsgupta]

This is unclaimed with no PR. Heads-up: the file/line moved — the bug is now in cli/eval_model.py (the CLI was refactored out of evaluate.py). model.eval() returns a dict {metrics_df, metrics_factory} for detection but a DataFrame for other tasks, so .to_csv() crashes for detection. I'll extract metrics_df using the same isinstance(..., dict) + .get("metrics_df") pattern already in cli/batch.py, and add the os.makedirs(...) that cli/eval_preds.py has. Opening a PR.