fix(sentiment): frequent "face not detected" results

[jvJUCA]

We often get a "no faces detected" response from the backend, even when the video clearly shows a face.

Logs from sentiment-analysis-api

2026-01-27 14:26:14.114 BRT
2026-01-27 17:26:14 - services.emotion_analysis.emotion_analysis_imp - INFO - Total frames in video: 476
2026-01-27 14:26:14.114 BRT
2026-01-27 17:26:14 - services.emotion_analysis.emotion_analysis_imp - INFO - Frames actually processed: 0
2026-01-27 14:26:14.114 BRT
2026-01-27 17:26:14 - services.emotion_analysis.emotion_analysis_imp - INFO - Total faces detected: 0
2026-01-27 14:26:14.114 BRT
2026-01-27 17:26:14 - services.emotion_analysis.emotion_analysis_imp - WARNING - No faces detected or no predictions made.
2026-01-27 14:26:14.114 BRT
2026-01-27 17:26:14 - services.emotion_analysis.emotion_analysis_imp - INFO - Percentages of emotions detected: {'Angry': 0, 'Disgusted': 0, 'Fearful': 0, 'Happy': 0, 'Neutral': 0, 'Sad': 0, 'Surprised': 0}
2026-01-27 14:26:14.115 BRT
2026-01-27 17:26:14 - routes.video_routes - INFO - Emotion analysis result: Angry=0.0 Disgusted=0.0 Fearful=0.0 Happy=0.0 Neutral=0.0 Sad=0.0 Surprised=0.0
2026-01-27 14:26:14.115 BRT
2026-01-27 17:26:14 - routes.video_routes - INFO - Time taken for analysis: 1.2495639324188232 seconds

[rahul810050]

@jvJUCA I will work on it and find the best solution...
/assign

[midaa1]

check this pr for solving :
#5

[MariamSalem4115]

I'm researching this issue for my GSoC 2026 proposal. I'm looking into how we can improve face detection for low-res videos, perhaps by integrating a fallback detector or a preprocessing layer.
@Aditi0522, I'd love to test your PR #15 with some edge-case videos

[Aditi0522]

@MariamSalem4115 Sure go ahead! I would be glad to know the results too, as to how my current model is performing in those edge case videos so that I can improve further.