World Alignment: unpaired cross-modal correspondence by relational identifiability

Method: scene states are sets of part states; relation fields are built
within each modality and are invariant to how each side labels its own
features; the cross-modal bridge is a coupling searched under an energy
that is a closed-form functional of one matrix; solving is spectral
initialisation followed by exact local refinement.

Evidence: in a procedurally generated closed world, blind recovery of a
hidden image-caption correspondence reaches 95.3% at 256 scenes against
0.39% chance, and the recovered pairs transfer to 200 held-out scenes at
93.0% exact retrieval with random-pair and shuffled-image controls at or
near chance. Cross-modal value correspondence is derived from disjoint
corpora rather than declared. On Visual Genome the field correlation
reaches 0.656 against the 0.9 that polynomial recovery needs, with the
deficit attributed away from segmentation and discretisation.

Protocol: no image-text pair enters any objective, optimiser,
initialisation, or model selection; hidden pairs score orderings only.

Co-Authored-By: Claude <noreply@anthropic.com>

1 files changed, 322 insertions, 0 deletions

diff --git a/artifacts/eval_paired.json b/artifacts/eval_paired.json
new file mode 100644
index 0000000..dbff1af
--- /dev/null
+++ b/artifacts/eval_paired.json
@@ -0,0 +1,322 @@
+{
+  "split": "test",
+  "samples": 1000,
+  "bridge_mode": "paired",
+  "retrieval": {
+    "i2t_r@1": 0.13699999451637268,
+    "t2i_r@1": 0.01899999938905239,
+    "i2t_r@5": 0.3540000021457672,
+    "t2i_r@5": 0.09799999743700027,
+    "i2t_r@10": 0.46399998664855957,
+    "t2i_r@10": 0.15600000321865082,
+    "i2t_median_rank": 13.0,
+    "t2i_median_rank": 68.0,
+    "chance_r@1": 0.001
+  },
+  "generation": {
+    "samples": 100,
+    "mean_best_reference_unigram_f1": 0.29334080910001414,
+    "examples": [
+      {
+        "row": 25,
+        "generated": "A man in a red shirt has a pair of sunglasses. He is wearing a red shirt. He is wearing a pair of sunglasses. He is wearing a red",
+        "references": [
+          "The man with pierced ears is wearing glasses and an orange hat.",
+          "A man with glasses is wearing a beer can crocheted hat.",
+          "A man with gauges and glasses is wearing a Blitz hat.",
+          "A man in an orange hat starring at something.",
+          "A man wears an orange hat and glasses."
+        ],
+        "unigram_f1": 0.25000000000000006
+      },
+      {
+        "row": 34,
+        "generated": "Black and white dogs chasing a brown dog with a green ball in a yard. 别墅男孩和狗追逐一个棕色的狗，用绿色的球",
+        "references": [
+          "A black and white dog is running in a grassy garden surrounded by a white fence.",
+          "A Boston Terrier is running on lush green grass in front of a white fence.",
+          "A black and white dog is running through the grass.",
+          "A dog runs on the green grass near a wooden fence.",
+          "A Boston terrier is running in the grass."
+        ],
+        "unigram_f1": 0.5161290322580646
+      },
+      {
+        "row": 51,
+        "generated": "Two students are practicing karate. One is practicing the front kick while the other is practicing the back kick. \nDo you want to practice? \nDo you",
+        "references": [
+          "A young female student performing a downward kick to break a board held by her Karate instructor.",
+          "Girl about to kick a piece of wood in half while karate instructor holds it",
+          "A girl kicking a stick that a man is holding in tae kwon do class.",
+          "A girl in karate uniform breaking a stick with a front kick.",
+          "A girl breaking boards by using karate."
+        ],
+        "unigram_f1": 0.1951219512195122
+      },
+      {
+        "row": 89,
+        "generated": "Two snowmobiles and a person wearing a winter coat and helmet. They are all dressed in winter clothing. \n\"\"\"\n    # Convert the string to a list",
+        "references": [
+          "Five snowmobile riders all wearing helmets and goggles line up in a snowy clearing in a forest in front of their snowmobiles; they are all wearing black snow pants and from left to right they are wearing a black coat, white coat, red coat, blue coat, and black coat.",
+          "Five people wearing winter jackets and helmets stand in the snow, with snowmobiles in the background.",
+          "Five people wearing winter clothing, helmets, and ski goggles stand outside in the snow.",
+          "A group of snowmobile riders gather in the snow.",
+          "Group gathered to go snowmobiling."
+        ],
+        "unigram_f1": 0.35616438356164387
+      },
+      {
+        "row": 97,
+        "generated": "A man is climbing up a tree. He is using a ladder.",
+        "references": [
+          "Two men sitting on the roof of a house while another one stands on a ladder.",
+          "Two men on a rooftop while another man stands atop a ladder watching them",
+          "Three men, one on a ladder, work on a roof.",
+          "People are fixing the roof of a house.",
+          "Three men are working on a roof."
+        ],
+        "unigram_f1": 0.30769230769230765
+      },
+      {
+        "row": 124,
+        "generated": "The couple is taking a picture in front of a group of people in a pool while the rest of the people watch.  From the left is a man in",
+        "references": [
+          "A bride in a light pink dress poses for a picture with male relatives and is being photographed by a man in a cream shirt with white pants.",
+          "A man in light colored clothing photographs a group of men wearing dark suits and hats standing around a woman dressed in a strapless gown.",
+          "A photographer takes a picture of a group of one girl in a pink dress and 10 boys in suits and hats.",
+          "A man photographs a woman in a pink dress and a throng of men in suits.",
+          "Man taking a photograph of a well dressed group of teens."
+        ],
+        "unigram_f1": 0.3846153846153846
+      },
+      {
+        "row": 133,
+        "generated": "Three people and their families are outside. They are dressed in winter clothing and shovel snow. They are outside in the snow.",
+        "references": [
+          "A group of 11 people in winter wear such as beanies, skiing jackets, gloves and backpacks are standing in snow paddles outside a house made of ice blocks while a person in front of the door seems to be leading them.",
+          "A group of people wearing snowshoes, and dressed for winter hiking, is standing in front of a building that looks like it's made of blocks of ice.",
+          "The people are quietly listening while the story of the ice cabin was explained to them.",
+          "A group of people standing in front of an igloo.",
+          "Several students waiting outside an igloo."
+        ],
+        "unigram_f1": 0.28571428571428575
+      },
+      {
+        "row": 134,
+        "generated": "Batter swinging a baseball with the catcher in the outfield and a player in the infield.",
+        "references": [
+          "A baseball player in red leaps into the air to avoid a player in blue, who is shortly behind him as he catches the ball, while other blue team members look on in the background.",
+          "A boy in a red uniform is attempting to avoid getting out at home plate, while the catcher in the blue uniform is attempting to catch him.",
+          "A baseball player coming into home plate with knee held high and the catcher right in front of him waiting for the ball.",
+          "A baseball player attempting to score a run before the catcher tags the player out.",
+          "A baseball catcher trying to tag a base runner in a baseball game."
+        ],
+        "unigram_f1": 0.46153846153846156
+      },
+      {
+        "row": 137,
+        "generated": "Three men standing on a brick wall while one of them is leaning on a pole.",
+        "references": [
+          "A man wearing a hat and a white shirt is cleaning windows.",
+          "A man in a white shirt stands high up on scaffolding",
+          "A man stands on boards on top of a huge ladder.",
+          "Man works on top of scaffolding.",
+          "A guy works on a building."
+        ],
+        "unigram_f1": 0.3703703703703703
+      },
+      {
+        "row": 199,
+        "generated": "person_sittingonthegroundwithabagofsandwichesandachair.",
+        "references": [
+          "A man wearing a reflective vest sits on the sidewalk and holds up pamphlets with bicycles on the cover.",
+          "Man in bright yellow vest displays bicycle safety information on street.",
+          "A man and woman talking are interrupted and a man covers his face.",
+          "A man in a vest is sitting in a chair and holding magazines.",
+          "A construction worker holds a sign about bicycling."
+        ],
+        "unigram_f1": 0.0
+      },
+      {
+        "row": 219,
+        "generated": "Two little boys hold each other at the moment of happiness.",
+        "references": [
+          "A mother decides to take her child on a piggyback ride outside their apartment complex.",
+          "A baby boy in a blue and white striped shirt is sitting on his mother's shoulders.",
+          "A mother and her young song enjoying a beautiful day outside.",
+          "A young woman is giving a baby a ride on her shoulders.",
+          "A woman gives a small child a piggyback ride."
+        ],
+        "unigram_f1": 0.0
+      },
+      {
+        "row": 222,
+        "generated": "Some men play basketball in the air while others jump in the water. People are not allowed to play basketball in the water.  Moderator:  Does this look",
+        "references": [
+          "A male volleyball player in red trunks tries to block the shot of the opposing player during a professional volleyball game.",
+          "Men playing volleyball, with one player missing the ball but hands still in the air.",
+          "A group of spectators watch a men's sand volleyball game.",
+          "Volleyball players compete on a sunny day at the beach.",
+          "Men playing volleyball in the sand."
+        ],
+        "unigram_f1": 0.23809523809523808
+      },
+      {
+        "row": 228,
+        "generated": "Stacy's mom is in a bakery and is selling cookies, she is wearing a apron and a chef hat. The man behind her is holding a tray",
+        "references": [
+          "A woman wearing a pink shirt and red apron stands in her restaurant holding food and surrounded by bagged breads, a microwave and the chalkboard menu up on the wall.",
+          "Woman wearing a yellow hat, pink shirt, and red apron is holding food in a kitchen.",
+          "Employee in pink shirt smiles for camera inside a restaurant kitchen.",
+          "A woman holding a bowl of food in a kitchen.",
+          "Women in store happily holding some food."
+        ],
+        "unigram_f1": 0.40909090909090917
+      },
+      {
+        "row": 246,
+        "generated": "Heulton, a 40-year-old man, works in a computer company. Heulton, a 40-year-old man, works in a computer",
+        "references": [
+          "An Asian man in spectacles and a t-shirt preparing some food.",
+          "A man working diligently while sitting at a table.",
+          "Man sitting using tool at a table in his home.",
+          "Indian man whittles in his own home.",
+          "A man is sharpening a knife."
+        ],
+        "unigram_f1": 0.22222222222222224
+      },
+      {
+        "row": 259,
+        "generated": "Two people in a rock wall climbing. They are surrounded by snow. They are wearing red and white. They are holding rocks. They are looking up. They",
+        "references": [
+          "Spelunkers pose inside a rock cavern while bathed in sunlight from the surface.",
+          "Shaft of light in a cave shows three spelunkers",
+          "Three rock climbers are posing on big rocks.",
+          "People stand inside a rock dome.",
+          "Three people sit in a cave."
+        ],
+        "unigram_f1": 0.1818181818181818
+      },
+      {
+        "row": 279,
+        "generated": "A young girl in a beautiful dress walks down the aisle of a store. The store is a toy store. The store is a toy store. The store is",
+        "references": [
+          "A girl in a polka dotted blue jean dress walks barefoot on a balance beam.",
+          "A girl in a jean dress is walking along a raised balance beam.",
+          "The little girl is walking a beam in gymnastics class.",
+          "A little girl balances on a gymnastics beam.",
+          "A little girl is walking on a balance beam."
+        ],
+        "unigram_f1": 0.3333333333333333
+      },
+      {
+        "row": 304,
+        "generated": "There is a man standing in the water, while the other man is wearing a helmet and standing with his feet in the water. - Wet. - Wet.",
+        "references": [
+          "A couple is sitting on the sand with their feet in the water, and they are shaking hands.",
+          "A man and woman wearing sunglasses sit halfway in the water.",
+          "A boy and a girl holding hands while sitting on a shore.",
+          "A couple with sunglasses are laying on the beach.",
+          "A blond holding hands with a guy in the sand."
+        ],
+        "unigram_f1": 0.40909090909090906
+      },
+      {
+        "row": 318,
+        "generated": "A woman in a grocery store aisle is looking through a rack of cosmetics. She's wearing a white shirt and black loafers. Behind her is a man in",
+        "references": [
+          "A woman standing with 3 other people in a store with two tables, some shelves with coffee and tea for sale, and a refrigerated drink case.",
+          "A woman in a gray sweater and black baseball cap is standing in line at a shop.",
+          "A woman in a hat waits to be served at a store.",
+          "Young people are standing in the store.",
+          "A woman in a cap at a coffee shop."
+        ],
+        "unigram_f1": 0.391304347826087
+      },
+      {
+        "row": 340,
+        "generated": "A person on a cliff face is walking up a slope. The cliff face is covered in green vegetation. \n\"\"\"\n\n# Part 2: Identify the type of",
+        "references": [
+          "The person has a striped shirt on and is holding on to a rope on a mountain.",
+          "A kid rock climbing against the backdrop of a green valley",
+          "The person in the striped shirt is mountain climbing.",
+          "A woman in a striped shirt climbs up a mountain.",
+          "A young man climbs a rocky hill."
+        ],
+        "unigram_f1": 0.3333333333333333
+      },
+      {
+        "row": 395,
+        "generated": "A person walks through a crowded street in a shopping district where a small, colorful, and eclectic shop is located in front of a row of colorful, vintage,",
+        "references": [
+          "Two young attractive women in front of a store drop coins for street performers.",
+          "Two women enjoy the entertainment of street performers.",
+          "Two men pretend to be statutes while women look on.",
+          "Two young women stop and watch street performers.",
+          "Pedestrians interact with street artists."
+        ],
+        "unigram_f1": 0.2380952380952381
+      },
+      {
+        "row": 434,
+        "generated": "Three men and two boys are standing together outside of the house. The men are all talking and the boys are standing behind them. The boys are holding up their",
+        "references": [
+          "A group of people standing on the lawn in front of a building.",
+          "Many people in blue jeans stand in front of a white church.",
+          "A large group of people stand outside of a church.",
+          "Family members standing outside a home.",
+          "People standing outside of a building."
+        ],
+        "unigram_f1": 0.16666666666666669
+      },
+      {
+        "row": 472,
+        "generated": "Two girls are playing soccer and one is holding a yellow and red soccer stick.",
+        "references": [
+          "A young girl wearing a blue shirt marching in a band playing a trumpet",
+          "Girl wearing blue shirt and black shorts plays trumped outside.",
+          "A teenager plays her trumpet on the field at a game.",
+          "A girl playing trumpet in a marching band.",
+          "Girl playing the trumpet in a marching band"
+        ],
+        "unigram_f1": 0.1739130434782609
+      },
+      {
+        "row": 487,
+        "generated": "A girl jumping off a diving board at the beach. on a jump.",
+        "references": [
+          "Person in a brown sweatshirt does a flip in midair over a trampoline with a river and electrical lines in the background.",
+          "A young adult is doing a back flip on a trampoline near a lake.",
+          "A woman is doing flips on a trampoline in front of a lake.",
+          "A woman does a somersault on a trampoline on the beach.",
+          "Trampolines are a fun way to exercise."
+        ],
+        "unigram_f1": 0.4999999999999999
+      },
+      {
+        "row": 520,
+        "generated": "一台电脑前有三个发光的LED灯。 一台电脑前有三个发光的LED灯。 一台电脑前有三个发光的LED灯。",
+        "references": [
+          "A man standing in a bar with a neon beer sign hanging on the wall",
+          "A man stands next to three video game machines and a beer sign.",
+          "A man stands next to a bank of computer gambling machines.",
+          "A man is standing by a group of video games in a bar.",
+          "A man stands next to three video machines."
+        ],
+        "unigram_f1": 0.0
+      },
+      {
+        "row": 539,
+        "generated": "A woman in a black dress and sunglasses uses a camera to take pictures of a camera. She is wearing a blue shirt and a black tank top. She is",
+        "references": [
+          "A woman in a green tank top looking at a drill while a crowd looks on.",
+          "A woman using a power drill in front of a group of people.",
+          "A woman uses a drill while another man takes her picture.",
+          "A woman, in a teal tank top, holding a power drill.",
+          "Spectators about to watch a woman use a drill."
+        ],
+        "unigram_f1": 0.3478260869565218
+      }
+    ]
+  }
+}
\ No newline at end of file