Ground results in the paper tables and rebuttal; drop parameter counts

README.md

@@ -22,15 +22,16 @@ View-invariant Video Representation Learning — **EMNLP 2026 (Main Conference)*
 
 [Paper][paper] · [Project page][project] · [Code][code]
 
-PRISM encodes a video clip into a **view-invariant** embedding: cosine similarity between
-two PRISM embeddings measures *what is happening*, not *where the camera is*. It gets there
-by refusing to produce a single embedding — each clip is decomposed into a view-invariant
-latent `z_vi` and a view-variant latent `z_vv`, and the split is enforced by recomposing
-across clips (`z_vi` of clip A with `z_vv` of clip B) and aligning the result with the
-matching recomposed language description. Semantics that leaked between the two streams
-cannot survive that swap.
+PRISM decomposes each video clip into a view-invariant latent `z_vi` and a view-variant
+latent `z_vv`, then recomposes them under language supervision — `z_vi` of clip A with
+`z_vv` of clip B is trained to match the embedding of the correspondingly recombined
+caption. Because that composition never occurs in the training data, semantics that leaked
+between the two streams cannot survive it. The clip embedding this model returns is the
+view-invariant stream, so cosine similarity measures *what is happening*, not *where the
+camera is*.
 
-This checkpoint is trained on **EgoExo4D**.
+This checkpoint is trained on **EgoExo4D**, with ego and exo views and captions generated
+by Gemini 3.
 
 ## Usage
 
@@ -65,17 +66,37 @@ For batched encoding of a clip manifest, see `scripts/encode.sh` in the [code re
 
 ## Results
 
-Cross-view semantic alignment (R@5 / top-1 / avg):
+**Cross-view semantic alignment.** Gains over the best baseline, ViewpointRosetta:
+**+10.4** Retrieval and **+11.5** Association, plus +7.46 Recognition and +7.32
+Anticipation. On Skill Assessment PRISM is comparable (55.28 vs 55.82) — proficiency cues
+depend on execution style rather than action identity, and land in the view-variant stream.
 
-| Method | <sub>EgoExo4D</sub><br>ego→exo | <br>exo→ego | <br>avg | <br>Recog. top-1 | <sub>EgoExoLearn</sub><br>Assoc. avg |
+| Method | <sub>**EgoExo4D**</sub><br>Retr. ego→exo | <br>exo→ego | <br>avg | <br>Recog. top-1 | <br>Skill | <sub>**EgoExoLearn**</sub><br>Assoc. avg | <br>Antic. avg | <br>Skill |
+|:---|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|
+| CLIP | 19.11 | 12.24 | 15.68 | 10.49 | 54.93 | 15.82 | 38.70 | 73.48 |
+| SigLIP2 | 35.08 | 19.72 | 27.40 | 13.86 | 55.57 | 26.6 | 64.60 | **76.03** |
+| LaViLa | 34.91 | 12.02 | 23.47 | 26.43 | 54.10 | 27.20 | 62.83 | 68.44 |
+| SUM-L | 47.14 | 32.77 | 39.96 | 24.83 | 55.10 | 4.64 | 45.50 | 65.31 |
+| ViewpointRosetta | 58.14 | 47.21 | 52.68 | 34.47 | **55.82** | 32.32 | 62.14 | 73.70 |
+| **PRISM** | **75.89** | **50.27** | **63.08** | **41.93** | 55.28 | **43.86** | **69.46** | 68.53 |
+
+**Fine-grained temporal modeling (AE2).** Best among out-of-domain models on all four
+tasks, and ahead of the best in-domain model on phase ordering and phase progression.
+
+| Method | AE2 videos | Frame retr.<br><sub>mAP@10</sub> | Phase order<br><sub>Kendall's τ</sub> | Phase class.<br><sub>F1</sub> | Phase prog.<br><sub>R²</sub> |
 |:---|:--:|:--:|:--:|:--:|:--:|
-| SigLIP2 | 35.1 | 19.7 | 27.4 | 13.9 | 26.6 |
-| ViewpointRosetta <sub>*prior SoTA*</sub> | 58.1 | 47.2 | 52.7 | 34.5 | 32.3 |
-| **PRISM** | **75.9** | **50.3** | **63.1** | **41.9** | **43.9** |
+| GTA | ✔ | 68.08 | 0.464 | 67.77 | 0.322 |
+| AE2 | ✔ | 73.20 | 0.562 | 74.47 | 0.480 |
+| SigLIP2 | ✘ | 45.56 | 0.020 | 43.91 | −1.322 |
+| ViewpointRosetta | ✘ | 54.17 | 0.047 | 46.93 | −0.150 |
+| **PRISM** | ✘ | **70.53** | **0.601** | **73.57** | **0.647** |
+
+**Robustness to background correlation (UNSCENE)** — videos whose action contradicts the
+background, averaged over three text encoders: PRISM reaches **14.9** R@10 / **0.181** RSA
+against 7.5 / 0.098 for ViewpointRosetta and 14.1 / 0.146 for DINOv2.
 
-Zero-shot transfer to AE2 (never trained on AE2 videos) beats AE2-trained models on
-temporal ordering — Kendall's τ **0.601** vs 0.562, phase progression **0.647** vs 0.480.
-Full tables and ablations are in the [paper][paper].
+Full tables, ablations, and the DEVIAS stream-probe analysis are in the [paper][paper] and
+the [code repo][code].
 
 ## Architecture
 
@@ -88,9 +109,9 @@ Full tables and ablations are in the [paper][paper].
 | Target encoder θ̄ | EMA of θ, decay 0.998 |
 | Embedding dim | 512 |
 
-This repo holds **trained weights only** (0.97 GB, fp32): θ, φ, θ̄, and the logit scale. The
-two backbones are re-downloaded from the Hub when the model is constructed, so nothing about
-them is duplicated here.
+This repo holds trained weights only: θ, φ, θ̄, and the logit scale. The two backbones are
+re-downloaded from the Hub when the model is constructed, so nothing about them is
+duplicated here.
 
 ## Training