wrathofgod/movie-scene-module0-v2

Module 0 Scene Encoder (v2)

Joint pretraining for movie-scene understanding with:

• supervised contrastive learning (temperature warmup + 80-negative effective batch)
• masked language modelling on annotation scene_text
• auxiliary scene-label prediction (SmoothL1 regression, label-smoothed classification)

Backbone

• base_model: microsoft/deberta-v3-base
• embedding_dim: 256

Dataset Summary

• all_scenes: 11118
• train_scenes: 9593
• val_scenes: 1525
• all_films: 80
• train_positive_avg_candidates: 6.0
• val_positive_avg_candidates: 6.0
• mlm_chunks: 15118

Best Validation Metrics

• val_loss: 3.6035
• val_contrastive_loss: 3.0713
• val_aux_loss: 0.8188
• val_r_at_1: 0.0766
• val_temperature: 0.1349
• avg_cls_acc: 0.5151
• avg_reg_mae: 0.9068
• avg_bin_acc: 0.7234
• val_task/emotional_valence: 1.1655
• val_task/scene_interaction_tone: 1.2965
• val_task/conflict_nature: 1.3281
• val_task/acoustic_space: 1.6019
• val_task/reality_layer: 0.3974
• val_task/score_dynamic_shape: 1.1894
• val_task/narrative_arc_position: 1.1239
• val_task/foreshadowing_type: 1.2793
• val_task/transition_type: 1.2792
• val_task/scene_tension_raw: 0.5380
• val_task/scene_arousal: 0.4447
• val_task/scene_valence_continuous: 0.4835
• val_task/pacing_intensity: 0.4516
• val_task/action_intensity: 0.5817
• val_task/emotional_shift_trigger: 0.5833
• val_task/emotion_tags: 0.4391

Intended Use

Use this encoder as the backbone for Module 1 by setting:

CFG["backbone"] = "/kaggle/working/module0_backbone"