Datasets:

Aymen-Lassoued
/

ABACUS_Bench

id stringlengths 22 34	image_id int64 716 581k	coco_split stringclasses 2 values	prompt stringlengths 22 35	count int64 1 14	category stringclasses 75 values
add_000000567118_traffic_light_K13	567,118	train	Add 13 traffic lights to the scene.	13	traffic light
add_000000302035_bench_K6	302,035	train	Add 6 benches to the scene.	6	bench
add_000000546171_elephant_K1	546,171	train	Add an elephant to the scene.	1	elephant
add_000000147526_parking_meter_K5	147,526	train	Add 5 parking meters to the scene.	5	parking meter
add_000000505945_cat_K4	505,945	train	Add 4 cats to the scene.	4	cat
add_000000291599_horse_K1	291,599	train	Add a horse to the scene.	1	horse
add_000000208350_potted_plant_K2	208,350	train	Add 2 potted plants to the scene.	2	potted plant
add_000000524029_bench_K1	524,029	train	Add a bench to the scene.	1	bench
add_000000485680_stop_sign_K1	485,680	train	Add a stop sign to the scene.	1	stop sign
add_000000080303_truck_K3	80,303	train	Add 3 trucks to the scene.	3	truck
add_000000236111_suitcase_K2	236,111	train	Add 2 suitcases to the scene.	2	suitcase
add_000000272487_bench_K3	272,487	train	Add 3 benches to the scene.	3	bench
add_000000153460_laptop_K2	153,460	train	Add 2 laptops to the scene.	2	laptop
add_000000324200_sandwich_K6	324,200	train	Add 6 sandwiches to the scene.	6	sandwich
add_000000397685_toilet_K2	397,685	train	Add 2 toilets to the scene.	2	toilet
add_000000362370_sheep_K4	362,370	train	Add 4 sheeps to the scene.	4	sheep
add_000000505945_cat_K2	505,945	train	Add 2 cats to the scene.	2	cat
add_000000028462_laptop_K1	28,462	train	Add a laptop to the scene.	1	laptop
add_000000147526_parking_meter_K4	147,526	train	Add 4 parking meters to the scene.	4	parking meter
add_000000177143_fork_K1	177,143	train	Add a fork to the scene.	1	fork
add_000000193629_boat_K1	193,629	train	Add a boat to the scene.	1	boat
add_000000498443_dog_K4	498,443	train	Add 4 dogs to the scene.	4	dog
add_000000262013_giraffe_K4	262,013	train	Add 4 giraffes to the scene.	4	giraffe
add_000000430325_stop_sign_K4	430,325	train	Add 4 stop signs to the scene.	4	stop sign
add_000000045359_vase_K10	45,359	train	Add 10 vases to the scene.	10	vase
add_000000395293_scissors_K6	395,293	train	Add 6 scissors to the scene.	6	scissors
add_000000357819_airplane_K2	357,819	train	Add 2 airplanes to the scene.	2	airplane
add_000000199487_baseball_bat_K1	199,487	train	Add a baseball bat to the scene.	1	baseball bat
add_000000083915_clock_K1	83,915	train	Add a clock to the scene.	1	clock
add_000000248276_donut_K5	248,276	train	Add 5 donuts to the scene.	5	donut
add_000000324893_sandwich_K1	324,893	train	Add a sandwich to the scene.	1	sandwich
add_000000334413_dog_K3	334,413	train	Add 3 dogs to the scene.	3	dog
add_000000038053_apple_K2	38,053	train	Add 2 apples to the scene.	2	apple
add_000000056557_zebra_K2	56,557	train	Add 2 zebras to the scene.	2	zebra
add_000000383270_cake_K3	383,270	train	Add 3 cakes to the scene.	3	cake
add_000000160933_pizza_K3	160,933	train	Add 3 pizzas to the scene.	3	pizza
add_000000182710_cell_phone_K1	182,710	train	Add a cell phone to the scene.	1	cell phone
add_000000385969_giraffe_K2	385,969	train	Add 2 giraffes to the scene.	2	giraffe
add_000000185153_bicycle_K1	185,153	train	Add a bicycle to the scene.	1	bicycle
add_000000177847_giraffe_K1	177,847	train	Add a giraffe to the scene.	1	giraffe
add_000000345497_carrot_K3	345,497	train	Add 3 carrots to the scene.	3	carrot
add_000000223754_potted_plant_K3	223,754	train	Add 3 potted plants to the scene.	3	potted plant
add_000000272487_bench_K2	272,487	train	Add 2 benches to the scene.	2	bench
add_000000430325_stop_sign_K7	430,325	train	Add 7 stop signs to the scene.	7	stop sign
add_000000383581_teddy_bear_K1	383,581	train	Add a teddy bear to the scene.	1	teddy bear
add_000000185591_bowl_K3	185,591	train	Add 3 bowls to the scene.	3	bowl
add_000000398385_traffic_light_K4	398,385	train	Add 4 traffic lights to the scene.	4	traffic light
add_000000420681_tv_K6	420,681	train	Add 6 tvs to the scene.	6	tv
add_000000145249_airplane_K2	145,249	train	Add 2 airplanes to the scene.	2	airplane
add_000000260917_horse_K1	260,917	train	Add a horse to the scene.	1	horse
add_000000151161_boat_K1	151,161	train	Add a boat to the scene.	1	boat
add_000000435433_dog_K3	435,433	train	Add 3 dogs to the scene.	3	dog
add_000000045359_vase_K8	45,359	train	Add 8 vases to the scene.	8	vase
add_000000420681_tv_K3	420,681	train	Add 3 tvs to the scene.	3	tv
add_000000475610_elephant_K3	475,610	train	Add 3 elephants to the scene.	3	elephant
add_000000431795_parking_meter_K2	431,795	train	Add 2 parking meters to the scene.	2	parking meter
add_000000577481_cow_K4	577,481	train	Add 4 cows to the scene.	4	cow
add_000000353559_bench_K2	353,559	train	Add 2 benches to the scene.	2	bench
add_000000418624_bus_K3	418,624	train	Add 3 buses to the scene.	3	bus
add_000000080303_truck_K8	80,303	train	Add 8 trucks to the scene.	8	truck
add_000000451667_parking_meter_K1	451,667	train	Add a parking meter to the scene.	1	parking meter
add_000000573572_airplane_K2	573,572	train	Add 2 airplanes to the scene.	2	airplane
add_000000379201_umbrella_K4	379,201	train	Add 4 umbrellas to the scene.	4	umbrella
add_000000095022_bird_K1	95,022	train	Add a bird to the scene.	1	bird
add_000000020979_motorcycle_K4	20,979	train	Add 4 motorcycles to the scene.	4	motorcycle
add_000000065371_cell_phone_K2	65,371	train	Add 2 cell phones to the scene.	2	cell phone
add_000000315607_elephant_K2	315,607	train	Add 2 elephants to the scene.	2	elephant
add_000000324200_sandwich_K11	324,200	train	Add 11 sandwiches to the scene.	11	sandwich
add_000000430325_stop_sign_K3	430,325	train	Add 3 stop signs to the scene.	3	stop sign
add_000000199410_giraffe_K2	199,410	train	Add 2 giraffes to the scene.	2	giraffe
add_000000170980_teddy_bear_K1	170,980	train	Add a teddy bear to the scene.	1	teddy bear
add_000000094828_giraffe_K8	94,828	train	Add 8 giraffes to the scene.	8	giraffe
add_000000044389_train_K1	44,389	train	Add a train to the scene.	1	train
add_000000346544_truck_K5	346,544	train	Add 5 trucks to the scene.	5	truck
add_000000395293_scissors_K4	395,293	train	Add 4 scissors to the scene.	4	scissors
add_000000368373_toilet_K2	368,373	train	Add 2 toilets to the scene.	2	toilet
add_000000018775_snowboard_K14	18,775	train	Add 14 snowboards to the scene.	14	snowboard
add_000000068625_giraffe_K6	68,625	train	Add 6 giraffes to the scene.	6	giraffe
add_000000202112_car_K2	202,112	train	Add 2 cars to the scene.	2	car
add_000000267649_giraffe_K2	267,649	train	Add 2 giraffes to the scene.	2	giraffe
add_000000127527_giraffe_K1	127,527	train	Add a giraffe to the scene.	1	giraffe
add_000000065371_cell_phone_K3	65,371	train	Add 3 cell phones to the scene.	3	cell phone
add_000000505945_cat_K6	505,945	train	Add 6 cats to the scene.	6	cat
add_000000184590_carrot_K3	184,590	train	Add 3 carrots to the scene.	3	carrot
add_000000153460_laptop_K1	153,460	train	Add a laptop to the scene.	1	laptop
add_000000192806_tie_K1	192,806	train	Add a tie to the scene.	1	tie
add_000000567118_traffic_light_K6	567,118	train	Add 6 traffic lights to the scene.	6	traffic light
add_000000132991_handbag_K1	132,991	train	Add a handbag to the scene.	1	handbag
add_000000347948_bus_K1	347,948	train	Add a bus to the scene.	1	bus
add_000000362159_traffic_light_K6	362,159	train	Add 6 traffic lights to the scene.	6	traffic light
add_000000036622_hot_dog_K1	36,622	train	Add a hot dog to the scene.	1	hot dog
add_000000368373_toilet_K3	368,373	train	Add 3 toilets to the scene.	3	toilet
add_000000176794_tennis_racket_K1	176,794	train	Add a tennis racket to the scene.	1	tennis racket
add_000000324200_sandwich_K4	324,200	train	Add 4 sandwiches to the scene.	4	sandwich
add_000000081571_giraffe_K1	81,571	train	Add a giraffe to the scene.	1	giraffe
add_000000074336_remote_K1	74,336	train	Add a remote to the scene.	1	remote
add_000000334954_surfboard_K3	334,954	train	Add 3 surfboards to the scene.	3	surfboard
add_000000080303_truck_K4	80,303	train	Add 4 trucks to the scene.	4	truck
add_000000125725_elephant_K6	125,725	train	Add 6 elephants to the scene.	6	elephant
add_000000218595_horse_K1	218,595	train	Add a horse to the scene.	1	horse

End of preview. Expand in Data Studio

ABACUS: A Benchmark for Image Editing Models to Assess Object Category and Counting Under Scene Conservation

ABACUS is a benchmark for numerical image editing: adding or removing an exact number of objects, or changing the category of an exact number of them, while preserving the rest of the scene. Evaluation is judge-free: each edit is verified at the instance level against COCONut panoptic ground truth confirmed by detector consensus, and no VLM or MLLM scores an edit or counts objects.

Code: github.com/AymenLass/ABACUS

At a glance

	Add	Remove	Change	Total
Easy	1,159	1,146	1,000	3,305
Medium	1,363	1,340	1,128	3,831
Hard	1,374	1,308	1,126	3,808
Total	3,896	3,794	3,254	10,944
max K	15	10	10

K is the requested count (the count field). The benchmark spans all 80 COCO thing-categories, with 72 change pairs.

Instruction formats

Config	Example
`add`	`Add 13 traffic lights to the scene.`
`remove`	`Delete 2 airplanes from the scene.`
`change`	`Change 2 airplanes into trains in the scene.`

Loading

from datasets import load_dataset

add    = load_dataset("Aymen-Lassoued/ABACUS_Bench", "add")                    # easy / medium / hard
remove = load_dataset("Aymen-Lassoued/ABACUS_Bench", "remove", split="hard")
change = load_dataset("Aymen-Lassoued/ABACUS_Bench", "change", split="medium")

There are three configs (add, remove, change), each with three splits (easy, medium, hard). The difficulty tier is the split name.

Data fields

Field	Configs	Description
`id`	all	Unique instruction id.
`image`	all	Source image to edit.
`image_id`	all	COCO 2017 image id.
`coco_split`	all	COCO 2017 split (`train` or `val`).
`prompt`	all	Edit instruction.
`count`	all	K, the number of objects to add / remove / change.
`category`	`add`, `remove`	Target category.
`source_category`	`change`	Category to be changed.
`target_category`	`change`	Category to change it into.

Evaluation

Each edit is scored as Score = CF · SC · OC (count fidelity, scene consistency, original-category conservation). The evaluation code and full protocol are available in the GitHub repository.

License

The benchmark instructions, counts and tier assignments are released under CC BY 4.0. Source images and panoptic labels come from COCO and COCONut under their respective licenses.

Citation

@misc{lassoued2026abacus,
  title  = {ABACUS: A Benchmark for Image Editing Models to Assess Object
            Category and Counting Under Scene Conservation},
  author = {Lassoued, Aymen and Loesch, Angelique and Audigier, Romaric},
  year   = {2026}
}

Downloads last month: 69