Datasets:

umd-zhou-lab
/

CoSTAR

image imagewidth (px) 768 1.28k	text stringlengths 17 330
	Replace the duck with a rabbit, recolor the castle to red, remove a bird, and estimate the depth of the scene. Also, remove the background and caption the image.
	Detect the children, add a cat near the child, recolor the slide to blue, remove the dog, and deblur the image while also captioning it.
	Replace the mirror with an abstract painting and recolor the cat to yellow. Detect a shoe rack in the image, add a potted plant, estimate the depth, and also segment the painting. Lastly, caption the image, and upscale it.
	Extend the image on the sides, segment the house, and estimate the depth of the final scene.
	Replace the lamp with a flower vase, recolor the table to black, remove the paintings on the wall all while upscaling the image and estimate its depth.
	Replace the balloon with a kite, segment the child, remove the cat, while debluring the image and estimating its depth.
	Remove the signboard, segment the animals, change the color of the elephant to yellow, add a person to the image, add a caption, and upscale the image as well.
	Replace the cutting board with a tray, recolor the tray to pink, remove the knife and add a bowl of fruit on the countertop. Also, remove the background, segment the tray, upscale the image, and estimate its depth.
	Replace the globe with a ball, recolor the ball to dark green, remove the chair, add a lamp on the ground, expand the image, segment the table, and upscale the image.
	Detect and segment the person, recolor dog to the yellow, detect the landmark and outpaint the image sideways. Also, add a hot air balloon in the sky, generate a caption for the image, and make it snowy
	Estimate the depth of the scene and give a nice caption for the image
	Modify the caption to “Bright Lights” and check if the new caption aligns with the image.
	Extend the image sideways and give a nice poetic caption for the final image.
	Detect the landmark and segment it, while also debluring the image.
	Detect the hikers, remove the dog, expand the image, and estimate the depth of the scene.
	Recolor the fountain to blue and remove the background
	Replace the boat with a yacht, segment the palm trees, detect the person, recolor the car to yellow, remove the rabbit, and caption the image.
	Replace the statue with a fountain, recolor the pedestal to gray, remove the trash bin, add a flower pot on the side, add a caption to the image, deblur the image, upscale the image, and change the scene to snowy.
	Replace the barn with a house, remove the cows, recolor the cock to a golden shade, and expand the image.
	Detect the landmark and replace the banner text with the name of the landmark + “Open”.
	Replace the laptop with a typewriter, remove the lamp, add a stack of books on the desk, remove the cat, segment the desk, and upscale the image while estimating its depth.
	Replace the “Closed” sign text with “Open” while detecting the pedestrian crossing and the trash bin on the corner, and also upscale the image.
	Detect the Landmark in the image
	Extract the quote and identify its author, analyze the sentiment of the quote, detect the bench, and recolor the bench to yellow and remove the cat.
	Detect the landmark and add the text “San Francisco Marathon 2024” on the banner.
	Replace the dog with a cat, remove the children in image, and also recolor the bench to yellow.
	Detect the family, segment the picnic table, replace the fruit basket with a hen, recolor the hen to blue, and remove the dog.
	Replace the flower vase with a lamp, recolor the plates to green, remove the painting on the wall, remove the dog, add a caption to the image. Also upscale the image while removing the background.
	Replace the lion with an elephant, recolor the zebra to pink, and caption the new image
	Detect the Eiffel Tower and add the text “Welcome to Paris” on the signboard. Also, provide a suitable caption for image.
	Remove the text and recolor the plane to yellow.
	Change the scenery to a bright sunny day.
	Detect the bridge and segment it. Remove the cars from the image and change the scenery to a clear night.
	Deblur the image.
	Transform the office desk into a dining table by replacing the laptop with a serving tray, the coffee mug with a wine glass, and the books with plates. Also, Replace the lamp with a flower vase, and remove the office chair.
	Update the classroom board by upscaling the image, detecting and extracting all text, redacting the text, analyzing the sentiment from the text, and detecting the board. Also, recolor the board to pink and genearte a caption for the image.
	Detect the sign, perform caption consistency check and then redact the text while replacing the cat with dog and recoloring the dog to pink.
	Replace the coffee cup with a teapot, recolor the table to blue, remove the flower vase, add a dessert plate, and also remove the background. Also, segment the teapot, upscale the image, and estimate the depth in image.
	Extract the label text, analyze its sentiment, replace the text with “Zero Sugar” all while detecting the can and upscaling the image.
	Detect the landmark, replace the text with name of landmark, recolor the car, and remove a trash can near the entrance.
	Remove the books and recolor the coffee cup to blue.
	Remove the pedestrians from the image and recolor the truck to green.
	Detect the cart and replace the sign with “Organic Smoothies”.
	Recolor the cows to blue, remove the fence from image and also estimate the depth of the image.
	Extract the text from the chalkboard, detect the children, analyze the sentiment of the quote, redact it, and recolor the board to dark green.
	Change the scenery to a bright sunny day and segment the sunflowers.
	Extract the caption, analyze its sentiment, check consistency of the caption, suggest a positive alternative caption while also detecting the phone screen, and upscaling the image.
	Recolor the cat to orange.
	Update the image by extracting the current text, replacing it with a modern quote, verifying that the new caption fits the scene, and analyzing the sentiment of the text while detecting the cat. Also, detect the new text and then redact it.
	Replace the umbrella with a dog, segment the boat, remove the people while also removing the background and upscaling the image.
	Detect the child, replace the ball with a kite and also expand the image on all sides
	Extract and redact the label text, and upscale the image.
	Recolor the table to yellow, detect a painting, remove a wall clock and enlarge the image. Also, estimate its depth, segment the table, give a caption, and upscale the image as well.
	Upscale the image and change the text to “Explore the Unknown”.
	Detect and recolor the car to yellow and add the text “For Sale” on the chalkboard.
	Detect the gazebo, add a fountain in between, recolor the gazebo to white, remove the duck, and upscale the image and give a nice caption for image.
	Refresh the coastal vista by detecting and identifying the landmark lighthouse, extracting the sign text, detecting the wooden sign, redacting the sign text, generating a nice caption for image, and upscaling the overall image.
	Remove the computer, recolor the desk to black, remove the desk lamp and add a coffee cup. Also, segment the desk, upscale the image, and expand it as well.
	Replace the lamp with a candle, remove the plant, segment the laptop, and recolor the book to green.
	Remove the dog, recolor the kite to blue, segment the child, and expand the image.
	Detect the Landmark in the photo and change the scenery to a sunny day.
	Replace the bicycle with an electric scooter, recolor it to yellow, remove a small trash can, add a flower pot next to the scooter, add a caption, make it snowy, and upscale the image.
	Detect the dog, segment the people, recolor the plants to red, and expand the image
	Replace the lamp with a candle, while extending the image and estimating the depth.
	Deblur the image, replace the text with “Explore”, detect the bird, and check if the new caption matches the image.
	Detect the deer, segment the elephant, recolor the deer to yellow while also removing the dog, and debluring the image.
	Extract the menu text and redact all prices greater than 2. Also, recolor a flower vase on the counter to red.
	Replace the boat with a cycle and remove the background from the image.
	Extract the event name and the remove the text.
	Detect the person, remove the dog, while upscaling the image and estimating the depth.
	Update the year on the billboard by extracting the current year and replacing it with “2025”. Also, verify caption consistency, analyze sentiment, redact old caption, and finally detect the plane flying. Finally, add a new caption on the top of image.
	Deblur the image and upscale it.
	Detect the giraffe in the image, segment the signboard, replace the hiker with a giraffe, remove the dog, and then upscale the image. Finally, give a caption to the image.
	Extract the tagline, perform sentiment analysis on the tagline and check if it aligns with the product image. Finally, also redact the text.
	Replace the ball with a frisbee, segment the dog, remove the background, and caption the image.
	Remove the umbrella, recolor the surfboard to red, remove a small beach towel, place a coconut drink on the sand, replace the cat with a rabbit, add a caption to the image, and then upscale and expand the image.
	Remove the bird while also removing the background from image.
	Remove the background, recolor the cat to blue, remove the people, and upscale the image.
	Remove the painting, recolor the bedsheets to pastel blue, detect the nightstand and add a floor lamp beside the bed. Also, remove the background from the image, segment the bed, and caption the image. Finally, upscale the image.
	Remove the background, recolor the flowers to pink shades, segment the flowers, and upscale the image.
	Detect and identify the building, replace the cat with a dog, recolor the dog to pink, and also segment it. Generate a caption for the image, and upscale it for clarity and expand the image vertically.
	Detect the signpost and rewrite the text as “Central Park”.
	Replace the coffee table with a glass one, recolor the sofa to navy blue, remove the painting, place a small potted plant on the table, add a caption to the image, deblur the image, and then upscale it.
	Deblur the image, upscale it, and detect the person in it.
	Replace the billboard text with “Grand Opening”, recolor the bird to pink, detect the bird, caption the image, and check if the new caption matches the image.
	Detect the statue, segment the cars, remove the dog, change the scene to evening tones, and also estimate the depth of the scene.
	Recolor the dog to yellow, the small birdhouse to yellow as well, and add a lantern hanging from the tree. Change the image to snowy, deblur it, upscale it, and expand it.
	Recolor the vase to red and segment the bouquet in the image.
	Extract the math equations and highlight the answer. Also, sharpen the image for clarity.
	Extract protest slogan, replace "Delayed" with "Now!," detect the protest sign and the people, check consistency of the new slogan, analyze sentiment, upscale the image.
	Analyze the concert banner by extracting texts, detecting the banner, checking if the banner text is consistent with the image, analyzing the sentiment, redacting the banner text, and upscaling the image. Also, write “BANNER” at top of the image.
	Detect the farmer, segment the plow, remove the buffalos, upscale the image, and caption it while estimating the depth.
	Replace the vase with a lamp and Estimate the depth.
	Replace the caravan with a truck, recolor the cacti to red, segment the truck while extending the desert and upscaling the image.
	Identify the landmark, remove the bird and add the landmark’s name to the signpost. Also, extract text from a banner, replace the “Welcome” text with “Hello,” and upscale the image.
	Detect the people in the restaurnt, extract the text on signage on table and redact it, detect the table cloth and recolor it to pink, generate a caption for the image and add it at the bottom and finally upsacle the image.
	Detect the wooden sign, redact the text all while replacing the cat with a dog, and checking the consistency of the caption (The text on the wooden sign).
	Analyze the park scene by detecting and extracting event details from a wooden sign, analyzing the sentiment, redacting the event details text, and upscaling the image for clarity. Also, detect the cat in the image and give a nice caption for the image.
	Replace the dog with a cat while removing the ball.
	Replace the farmhouse with a barn, recolor the cow to green, extend the image on all sides and segment the barn.

End of preview. Expand in Data Studio

🎨 CoSTA* Dataset

CoSTA* is a multimodal dataset for multi-turn image-to-image transformation tasks, designed to accompany the FaSTA* agent presented in FaSTA$^*$: Fast-Slow Toolpath Agent with Subroutine Mining for Efficient Multi-turn Image Editing. It provides:

High-quality images for various image editing tasks.
Detailed text prompts describing the desired transformations.
Multimodal tasks including inpainting, object recoloring, object segmentation, object replacement, text replacement, and more.

GitHub Repo

Try out the CoSTA* agent online: Live Demo

📂 Dataset Structure

metadata.csv → List of image-text pairs.
{file_name}.jpeg → All images in the main directory.
dataset_info.json → Hugging Face dataset metadata.

Downloads last month: 826

Size of downloaded dataset files:

63.8 MB

Size of the auto-converted Parquet files:

63.7 MB

Number of rows:

121