When the Codec Hallucinates: User Perceptions of Miscompressed Images - Userstudy Dataset

This dataset is released as part of the publication "When the Codec Hallucinates: User Perceptions of Miscompressed Images", accepted at the Conference on Human Factors in Computing Systems 2026 (ACM CHI) in Barcelona, Spain. Paper:https://informationsecurity.uibk.ac.at/pdfs/HB2026_CHI.pdf For better reproducibility, we provide both the dataset of images used for the study and the detailed responses of the survey. image_dataset.csv Provides detailed information for every image used in the study. Refer to Figure 4 in the paper for an overview; each box in the figure reflects one row in this table. Columns ID: Unique identifier for each image sample, consisting of [version|name|group] version: J - JPEG-compressed N - Neural-compressed M - Miscompressed U - Uncompressed name: Five first digits of image name group: Subject group that saw the image. [1...4], or 0 if all groups saw the image ("anchor image") ORIGNAME: Full name of the scene (original image) CROP: Size of the crop in pixels (quadratic) GROUP: Subject group that saw the image. [1...4], or 0 if all groups saw the image ("anchor image") ORDER: Position in the instrument [1...12] UNCOMP: "1" if the image is uncompressed JPEG: "1" if the image isas JPEG JPEGQ: JPEG quality factor for JPEG compressed images NC: "1" if the image is neurally compressed NCQ: Compression codec for neurally compressed images MISCOMP: "1" if the image contains a miscompression survey_responses.csv For the corresponding questions, please refer to Appendix B.2 of the paper. Columns represent questions in chronological order, followed by timings. Columns DEM- age, sex D1 - D3: demographic questions [version|name|group|question]: Image comparison version: J - JPEG-compressed N - Neural-compressed M - Miscompressed U - Uncompressed name: Five first digits of image name group: Subject group that saw the image [1...4], or 0 if all groups saw the image ("anchor image") question: dif - S1: difference mis - S2: misunderstandings [1...6], 1 = "certainly" mod - S3(1): intentional editing [1...6], 1 = "certainly" com - S3(2): uncontrolled distortion [1...6], 1 = "certainly" THE* C1(1 - 8): Theoretical knowledge PRA* C2(1 - 8): Practical experience VER* C3(1 - 6): Verification practices EXP* C4(1 - 2): Experience with image distribution interviewtime: Time to complete the whole survey [s] age, group, vis -Time: Demographics Time [s] [version|name|group|question] -Time: Image comparison time [s] ... question: dif mis modcom: mod and com together since on the same display the, pra, ver, exp -Time: Control variable's time [s]