Repository navigation
docs: note the scenario-views whose take 2 reuses take 1 - #86
Merged
rgeirhos merged 2 commits intoOct 7, 2026
Merged
Conversation
Add a README section listing the 40 scenario-views where take 2 is the take-1 recording, grouped by the reasons given in google-deepmind#84, with the effect on the MSE component of the verified per-view score and how Physics-IQ Verified differs. Link it from the "Variations" key feature.
rgeirhos
approved these changes
Oct 7, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
As discussed in #84 with @sam-motamed, this adds a short README section on the scenario-views whose take 2 reuses the take-1 recording.
How the statements were checked:
split-videos/testing/8FPS/andsplit-videos/testing/30FPS/(the script is in take-2 is a byte-identical copy of take-1 in 40 of the 198 test scenario-views #84).IQTablefromphysiq/calculate_iq_score_stable.pyat 266e5fc and the public per-view results of Kandinsky-WM 1.0 (four seeds, 24 FPS), the MSE component on the 36 scenario-views with zero take-to-take MSE equalsclip(1e-8 / model_mse, 0, 1)and stays below 3e-5, and the per-view score there is at most 0.750007. The full-credit case described in the README comes fromparse_list_of_floatsinphysiq/calculate_iq_score.py, which rounds each per-frame value to four decimals: a model MSE of zero gives1e-8 / 0, which is clipped to 1.b65f567, the take-1 and take-2 files of magnet-domino (left) and of the three match-blows-balloon views differ in the 8, 16, 24 and 30 FPS folders, and their take-to-take MSE at 24 FPS, as the evaluator reads it (per-frame values rounded to four decimals), ranges from 8.3e-5 to 3.45e-4. Among the other 36, duck-falls-in-box (center) is also a different file in every folder, and glass-stays-same in the 8 and 30 FPS folders; at 24 FPS their take-to-take MSE is zero after rounding.This is a documentation change only. No code or data is touched.