ropedia-xperience-10m-task-baselines / TASK_METHOD_20_RESULT_MATRIX.md
cy0307's picture
Add files using upload-large-folder tool
d053290 verified
|
Raw
History Blame
3.56 kB

Task Method 20-Result Matrix

Every method has one record for each of the 20 unified task contracts. Numeric scores appear only where a committed runner or verified package produced that task target.

Legend: score = direct numeric task score and proxy = documented compact substitute target. The current public matrix is complete at 180/180 scored records; unsupported/not-evaluated labels are retained only for future regression audits.

Method Records Scored Proxy scored Scoreless Status counts
Minimal 20 20 0 0 scored 20
Neural MLP 20 20 0 0 scored 20
128ep Aligned Simple 20 20 1 0 proxy scored 1, scored 19
128ep Aligned NN 20 20 1 0 proxy scored 1, scored 19
128ep Raw Simple 20 20 2 0 proxy scored 2, scored 18
128ep Raw NN 20 20 2 0 proxy scored 2, scored 18
Qwen3-Omni v6 LoRA 20 20 0 0 scored 20
Cosmos3-Super Reasoner 20 20 0 0 scored 20
Cosmos3-Nano Future Window 20 20 0 0 scored 20
# Task Min NN 128-S 128-NN 128-RS 128-RN Qwen3 C3-S C3-N
01 Action Recognition score score score score score score score score score
02 Procedure Step Recognition score score score score score score score score score
03 Action Boundary Detection score score score score score score score score score
04 Next-Action Prediction score score score score score score score score score
05 Hand Trajectory Forecasting score score score score score score score score score
06 Contact State Prediction score score score score score score score score score
07 Object Relevance Prediction score score score score score score score score score
08 Language Grounding score score score score score score score score score
09 Cross-Modal Retrieval score score score score score score score score score
10 Cross-Modal Reconstruction score score score score score score score score score
11 Temporal Order Verification score score score score score score score score score
12 Multimodal Synchronization Detection score score score score score score score score score
13 Long-Horizon Next-Action Forecasting score score score score score score score score score
14 Long-Horizon Next-Subtask Forecasting score score score score score score score score score
15 Interaction Text Prediction score score score score proxy proxy score score score
16 Action-Object Relation Prediction score score score score score score score score score
17 Future Object-Set Forecasting score score score score score score score score score
18 IMU-to-Hand Pose Reconstruction score score score score score score score score score
19 Camera-View Synchronization Retrieval score score proxy proxy proxy proxy score score score
20 Time-to-Next-Transition Regression score score score score score score score score score

Sources and raw values are in docs/data/task_method_20_result_matrix.json and docs/data/unified_task_model_radar.json.