Dataset Viewer
Auto-converted to Parquet Duplicate
image_id
int64
9
582k
image_path
stringlengths
41
41
caption
stringlengths
21
250
negative_caption
stringlengths
21
267
type
stringclasses
7 values
318,556
train2014/COCO_train2014_000000318556.jpg
A very clean and well decorated empty bathroom
A very clean and well decorated empty bed
REPLACE-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A very clean and well decorated empty bathroom
A very loud and well decorated empty bathroom
REPLACE-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
A very clean and well decorated empty bathroom
A very clean and well decorated empty bathroom and a garage
ADD-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A very clean and well decorated empty bathroom
A very clean and well decorated empty and messy bathroom
ADD-ATT
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A panoramic view inside a kitchen and all of its appliances.
REPLACE-REL
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A kitchen of a panoramic view and all of its appliances.
SWAP-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A panoramic painting of a kitchen and all of its appliances.
REPLACE-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A brilliant view of a kitchen and all of its appliances.
REPLACE-ATT
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A panoramic view and a section of a kitchen and all of its appliances.
ADD-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic view of a kitchen and all of its appliances.
A panoramic and classical view of a kitchen and all of its appliances.
ADD-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
A blue and white bathroom under butterfly themed wall tiles.
REPLACE-REL
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
Butterfly with a blue and white bathroom themed wall tiles.
SWAP-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
A blue and white restaurant with butterfly themed wall tiles.
REPLACE-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
A black and white bathroom with butterfly themed wall tiles.
REPLACE-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
A blue and white bathroom and a partial with butterfly themed wall tiles.
ADD-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A blue and white bathroom with butterfly themed wall tiles.
A blue and white bathroom with black butterfly themed wall tiles.
ADD-ATT
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A panoramic photo inside a kitchen and dining room
REPLACE-REL
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A kitchen and dining room of a panoramic photo
SWAP-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A panoramic painting of a kitchen and dining room
REPLACE-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A tiny photo of a kitchen and dining room
REPLACE-ATT
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A panoramic photo and an image of a kitchen and dining room
ADD-OBJ
116,100
train2014/COCO_train2014_000000116100.jpg
A panoramic photo of a kitchen and dining room
A panoramic and industrial photo of a kitchen and dining room
ADD-ATT
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
A graffiti-ed stop sign up the street from a red car
REPLACE-REL
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
The street across a graffiti-ed stop sign from a red car
SWAP-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
A graffiti-ed stop sign across the bridge from a red car
REPLACE-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
A graffiti-ed stop sign across the street from a subway car
REPLACE-ATT
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
A graffiti-ed stop sign and a silence across the street from a red car
ADD-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A graffiti-ed stop sign across the street from a red car
A graffiti-ed empty stop sign across the street from a red car
ADD-ATT
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and a red beetle near the road
REPLACE-REL
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and the road on a red beetle
SWAP-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A redd stop sign and a vandalize beetle on the road
SWAP-ATT
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and a red rose on the road
REPLACE-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and a black beetle on the road
REPLACE-ATT
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and a silence and a red beetle on the road
ADD-OBJ
379,340
train2014/COCO_train2014_000000379340.jpg
A vandalized stop sign and a red beetle on the road
A vandalized stop sign and a red and gold beetle on the road
ADD-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A bathroom without a border of butterflies and blue paint on the walls above it.
REPLACE-REL
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A border with a bathroom of butterflies and blue paint on the walls above it.
SWAP-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A couch with a border of butterflies and blue paint on the walls above it.
REPLACE-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A bathroom with a border of butterflies and black paint on the walls above it.
REPLACE-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A bathroom and a partial with a border of butterflies and blue paint on the walls above it.
ADD-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
A bathroom with a border of butterflies and blue paint on the walls above it.
A upper bathroom with a border of butterflies and blue paint on the walls above it.
ADD-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
An angled view of a beautifully decorated bathroom.
An angled view inside a beautifully decorated bathroom.
REPLACE-REL
318,556
train2014/COCO_train2014_000000318556.jpg
An angled view of a beautifully decorated bathroom.
An decorated view of a beautifully angled bathroom.
SWAP-ATT
318,556
train2014/COCO_train2014_000000318556.jpg
An angled view of a beautifully decorated bathroom.
An angled doorway of a beautifully decorated bathroom.
REPLACE-OBJ
318,556
train2014/COCO_train2014_000000318556.jpg
An angled view of a beautifully decorated bathroom.
An angled view and a thence of a beautifully decorated bathroom.
ADD-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
The two people are walking down the beach.
The two people are walking up the beach.
REPLACE-REL
134,754
train2014/COCO_train2014_000000134754.jpg
The two people are walking down the beach.
The beach are walking down the two people.
SWAP-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
The two people are walking down the beach.
The two kids are walking down the beach.
REPLACE-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
The two people are walking down the beach.
The two people and a world are walking down the beach.
ADD-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
The two people are walking down the beach.
The two fifty people are walking down the beach.
ADD-ATT
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A sink and a toilet beside a small bathroom.
REPLACE-REL
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A sink and a small bathroom inside a toilet.
SWAP-OBJ
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A bed and a toilet inside a small bathroom.
REPLACE-OBJ
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A sink and a toilet inside a large bathroom.
REPLACE-ATT
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A sink and a balance and a toilet inside a small bathroom.
ADD-OBJ
538,480
train2014/COCO_train2014_000000538480.jpg
A sink and a toilet inside a small bathroom.
A empty sink and a toilet inside a small bathroom.
ADD-ATT
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An empty kitchen without white and black appliances.
REPLACE-REL
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
White and black appliance with an empty kitchens.
SWAP-OBJ
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An white kitchen with empty and black appliances.
SWAP-ATT
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An empty bedroom with white and black appliances.
REPLACE-OBJ
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An elaborate kitchen with white and black appliances.
REPLACE-ATT
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An empty kitchen and a work with white and black appliances.
ADD-OBJ
476,220
train2014/COCO_train2014_000000476220.jpg
An empty kitchen with white and black appliances.
An empty and little kitchen with white and black appliances.
ADD-ATT
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
A white square kitchen under tile floor that needs repairs
REPLACE-REL
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
Tile floor with a white square kitchen that needs repairs
SWAP-OBJ
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
A white square roof with tile floor that needs repairs
REPLACE-OBJ
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
A round square kitchen with tile floor that needs repairs
REPLACE-ATT
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
A white square kitchen and a work with tile floor that needs repairs
ADD-OBJ
299,675
train2014/COCO_train2014_000000299675.jpg
A white square kitchen with tile floor that needs repairs
A white square kitchen with red tile floor that needs repairs
ADD-ATT
32,275
train2014/COCO_train2014_000000032275.jpg
The vanity contains two sinks with a towel for each.
The vanity faced two sinks with a towel for each.
REPLACE-REL
32,275
train2014/COCO_train2014_000000032275.jpg
The vanity contains two sinks with a towel for each.
Two sink contains the vanitys with a towel for each.
SWAP-OBJ
32,275
train2014/COCO_train2014_000000032275.jpg
The vanity contains two sinks with a towel for each.
The pool contains two sinks with a towel for each.
REPLACE-OBJ
32,275
train2014/COCO_train2014_000000032275.jpg
The vanity contains two sinks with a towel for each.
The vanity and a scandal contains two sinks with a towel for each.
ADD-OBJ
32,275
train2014/COCO_train2014_000000032275.jpg
The vanity contains two sinks with a towel for each.
The another vanity contains two sinks with a towel for each.
ADD-ATT
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
Several metal balls sit in the sand behind a group of people.
REPLACE-REL
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
The sands sit in several metal ball near a group of people.
SWAP-OBJ
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
Several metal balls sit in the water near a group of people.
REPLACE-OBJ
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
two metal balls sit in the sand near a group of people.
REPLACE-ATT
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
Several metal balls and a man sit in the sand near a group of people.
ADD-OBJ
302,443
train2014/COCO_train2014_000000302443.jpg
Several metal balls sit in the sand near a group of people.
several and fewer metal balls sit in the sand near a group of people.
ADD-ATT
134,754
train2014/COCO_train2014_000000134754.jpg
Two people carrying surf boards on a beach.
Two people kicking surf boards on a beach.
REPLACE-REL
134,754
train2014/COCO_train2014_000000134754.jpg
Two people carrying surf boards on a beach.
Surf board carrying two peoples on a beach.
SWAP-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
Two people carrying surf boards on a beach.
Two soldiers carrying surf boards on a beach.
REPLACE-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
Two people carrying surf boards on a beach.
Two people and a world carrying surf boards on a beach.
ADD-OBJ
134,754
train2014/COCO_train2014_000000134754.jpg
Two people carrying surf boards on a beach.
Two fifty people carrying surf boards on a beach.
ADD-ATT
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
A kitchen was brown cabinets, tile backsplash, and grey counters.
REPLACE-REL
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
Brown cabinet with a kitchens, tile backsplash, and grey counters.
SWAP-OBJ
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
A basement with brown cabinets, tile backsplash, and grey counters.
REPLACE-OBJ
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
A kitchen with black cabinets, tile backsplash, and grey counters.
REPLACE-ATT
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
A kitchen and a work with brown cabinets, tile backsplash, and grey counters.
ADD-OBJ
25,470
train2014/COCO_train2014_000000025470.jpg
A kitchen with brown cabinets, tile backsplash, and grey counters.
A empty kitchen with brown cabinets, tile backsplash, and grey counters.
ADD-ATT
513,461
train2014/COCO_train2014_000000513461.jpg
A surfer, a woman, and a child walk on the beach.
A soldier, a woman, and a child walk on the beach.
REPLACE-OBJ
513,461
train2014/COCO_train2014_000000513461.jpg
A surfer, a woman, and a child walk on the beach.
A plastic surfer, a woman, and a child walk on the beach.
ADD-ATT
18,691
train2014/COCO_train2014_000000018691.jpg
A few people sit on a dim transportation system.
A dim transportation system sit on a few people.
SWAP-OBJ
18,691
train2014/COCO_train2014_000000018691.jpg
A few people sit on a dim transportation system.
A few trucks sit on a dim transportation system.
REPLACE-OBJ
18,691
train2014/COCO_train2014_000000018691.jpg
A few people sit on a dim transportation system.
A few people sit on a modern transportation system.
REPLACE-ATT
18,691
train2014/COCO_train2014_000000018691.jpg
A few people sit on a dim transportation system.
A few people and a world sit on a dim transportation system.
ADD-OBJ
18,691
train2014/COCO_train2014_000000018691.jpg
A few people sit on a dim transportation system.
A few many people sit on a dim transportation system.
ADD-ATT
285,579
train2014/COCO_train2014_000000285579.jpg
A person, protected from the rain by their umbrella, walks down the road.
A person, protected with the rain by their umbrella, walks down the road.
REPLACE-REL
285,579
train2014/COCO_train2014_000000285579.jpg
A person, protected from the rain by their umbrella, walks down the road.
A person, protected from their umbrella by the rain, walks down the road.
SWAP-OBJ
End of preview. Expand in Data Studio

CS-CLIP data

Training data and benchmark files for CS-CLIP: Compositional Scene Graph-guided CLIP for Robust Compositional Reasoning (Findings of EMNLP 2026).

Images are not included. Download COCO, Visual Genome (ARO) and GQA images from their original sources; the code repository explains how.

Files

File Records Content
train/cs_clip_hard_negatives.jsonl 1,941,340 Caption / hard-negative pairs on 78,614 COCO train2014 images. This is the exact training set of the released model.
eval/gqa_two_choice.json 10,000 GQA two-choice benchmark built from the GQA balanced validation split.
eval/controlled_object.json 23,937 Controlled test cases: one object replaced (from VG-Relation).
eval/controlled_attribute.json 28,748 Controlled test cases: one attribute replaced (from VG-Attribute).
eval/controlled_relation.json 13,595 Controlled test cases: one relation replaced (from VG-Relation).

Training data

Each line is one hard negative:

{"image_id": 318556, "image_path": "train2014/COCO_train2014_000000318556.jpg",
 "caption": "A very clean and well decorated empty bathroom",
 "negative_caption": "A very clean and well decorated empty bed", "type": "REPLACE-OBJ"}
  • caption is a human-written COCO caption.
  • negative_caption changes one compositional element of the caption.
  • type gives the operation:
Type Count
REPLACE-OBJ 354,906
REPLACE-ATT 174,368
REPLACE-REL 311,156
SWAP-OBJ 318,529
SWAP-ATT 41,821
ADD-OBJ 365,583
ADD-ATT 374,977

How the negatives were made:

  1. Captions are parsed into scene graphs with SceneGraphParser.
  2. For Replace and Add, bert-base-uncased proposes candidates for the target span, and roberta-large-mnli keeps the most contradictory one.
  3. Swap is rule-based.

The negatives are generated automatically and are not human-verified.

  • A small fraction is ungrammatical, or differs from the caption only in letter case (about 4K SWAP-OBJ items).
  • Because the masked language model fills words freely, some negatives contain violent, sexual or profane words that are not in the original caption, for example "dead", "corpse", "naked" or "prostitute". This affects about 0.2% of pairs.
  • The negatives are meant as contrastive training signal. Do not use them as natural image descriptions.

Benchmark files

  • GQA two-choice. {"question_id", "image", "question", "positive", "negative"}. positive is the question followed by the gold full answer; negative replaces the answer with another entity from the same image's scene graph. image is relative to the GQA image folder.
  • Controlled test cases. {"image", "positive", "negative"}. They are built from the VG-Relation and VG-Attribute sets of ARO (Yuksekgonul et al., ICLR 2023). Each negative replaces exactly one object, attribute or relation of the positive caption with a word sampled from the Visual Genome vocabulary. image is relative to the ARO image folder (vgr_vga_images.zip). Models are evaluated on full images.

License

CC BY 4.0. Captions come from COCO and annotations from Visual Genome and GQA, all released under CC BY 4.0. The controlled test cases are derived from ARO's VG-Relation and VG-Attribute. Images are not included and remain subject to their original terms.

Citation

@inproceedings{jeong2026csclip,
  title     = {{CS-CLIP}: Compositional Scene Graph-guided {CLIP} for Robust Compositional Reasoning},
  author    = {Jeong, SeongJun and Jung, Minjoon and Choi, Woo Suk and Jang, Youwon and Zhang, Byoung-Tak},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2026},
  year      = {2026}
}
Downloads last month
-

Models trained or fine-tuned on juns/CS-CLIP-data

Paper for juns/CS-CLIP-data