Fix detect.py CSV output crashing on undefined NUM_OUTPUT and wrong df key

[Chessing234]

Bug

The CSV-output branch of python/detect.py's main() cannot run end-to-end: it raises NameError on NUM_OUTPUT and, even if that were defined, KeyError on df['feat'].

https://github.com/BVLC/caffe/blob/9b89154/python/detect.py#L157-L163

if args.output_file.lower().endswith('csv'):
    # csv
    # Enumerate the class probabilities.
    class_cols = ['class{}'.format(x) for x in range(NUM_OUTPUT)]
    df[class_cols] = pd.DataFrame(
        data=np.vstack(df['feat']), index=df.index, columns=class_cols)
    df.to_csv(args.output_file, cols=COORD_COLS + class_cols)

Root cause

• NUM_OUTPUT is referenced but never defined in the module (no import, no assignment).
• Detector.detect_windows builds each detection row with the key 'prediction' (see detector.py:93-97), so the resulting DataFrame has a prediction column — there is no feat column.

Fix

Stack df['prediction'] once, derive the number of output classes from its shape, and feed the already-stacked array into the per-class DataFrame. No more NUM_OUTPUT, no more df['feat'], and the number of columns now reflects the actual prediction vector. The h5 branch is untouched.