Multi-exposure image stacks for testing of HDR deghosting methods ================================================================= Copyright (c) 2016, Kanita Karaduzovic-Hadziabdic This dataset contains a set of RAW and JPEG image stacks of multi-exposure image sequences that could be used for testing HDR reconstruction and deghosting methods. The unique feature of this dataset is that it contains both test image stacks, with motion and misalignment, and motion-free reference image stacks, which can be used with full-reference image quality metrics. This dataset is priovided under the Creative Commons Attribution license (CC BY). If you use this dataset in your publication, please cite the following paper: Kanita Karaduzovic-Hadziabdic, Jasminka Hasic Telalovic, Rafal K. Mantiuk Assessment of multi-exposure HDR image deghosting methods In: Computers & Graphics, Volume 63, April 2017, Pages 1–17 http://dx.doi.org/10.1016/j.cag.2017.01.002 The complete dataset consist of 36 scenes, provided as: * RAW (Canon) image stacks * JPEG image stacks * Merged HDR images (using several methods) In addition to that, we also provide the results of the subjective quality assessment experiment peformed on the set of merged HDR images. Because of the size of the dataset, it has been split into the following .ZIP files: * exposure_stacks_part1.zip * exposure_stacks_part2.zip * exposure_stacks_part1.zip * merged_hdr_images_part1.zip * merged_hdr_images_part2.zip * merged_hdr_images_part3.zip * pairwise_comparison_results The 36 scenes are oranized into 9 categories of motion type, each containing 4 image sets. The categories are as follows: * complex motion : Highly dynamic scene with small/large motion displacement of small/large objects, non-rigid motion, occlusion, and several independently moving objects. handheld : Static scene captured with a handheld camera * lolm : Large object displacement with large motion. * losm : Large object displacement with small motion. multiview : Multi-view sequence of a static scene. * nrm : Motion of non-rigid and high texture objects. * occlusion : Scene containing occlusion. * solm : Small object displacement with large motion. * sosm : Small object displacement with small motion Scenes marked with * indicate dynamic scenes captured on a tripod where objects were moved between LDR image capture to simulate motion. For each scene, both test and reference multi-exposure sequence were captured. Test exposure sequence refers to the sequence of multi-exposures that contain either objects or camera motion. Reference exposure sequence refers to the sequence of multi-exposures where all pixels are perfectly aligned (i.e. ground truth sequence). ---------------------------------------------------------------------- --------------------- Exposure stacks -------------------------------- ---------------------------------------------------------------------- Exposure_stacks contain sequences of multi-exposure images, of both ghosted (with motion) and ground truth (perfectly aligned) images. NOTE: For exposure sequences with objects in motion, the position in the middle of the motion was selected for the reference exposure sequence. Because of this, in 'ghosted' folder within the exposure_stacks, the middle exposed images has 'gt' in the file name and it is the copy of the corresponding middle exposed ground truth image. The structure of exposure_stacks is organized as follows: exposure_stacks complex image_set1 ghosted jpg setup1_complex_1.jpg setup1_complex_2jpg setup1_complex_3gt3.jpg setup1_complex_4.jpg setup1_complex_5.jpg raw setup1_complex_1.cr2 setup1_complex_2.cr2 setup1_complex_3gt3.cr2 setup1_complex_4.cr2 setup1_complex_5.cr2 ground_truth jpg setup1_complex_gt1.jpg setup1_complex_gt2.jpg setup1_complex_gt3.jpg setup1_complex_gt4.jpg setup1_complex_gt5.jpg raw setup1_complex_gt1.cr2 setup1_complex_gt2.cr2 setup1_complex_gt3.cr2 setup1_complex_gt4.cr2 setup1_complex_gt5.cr2 image_set2 ghosted jpg setup2_complex_1.jpg setup2_complex_2.jpg ....... raw setup2_complex_1.cr2 setup2_complex_2.cr2 ....... ground_truth jpg setup2_complex_gt1.jpg setup2_complex_gt2.jpg ....... raw setup2_complex_gt1.cr2 setup2_complex_gt2.cr2 ....... image_set3 ...... image_set4 ...... handheld lolm losm ... ... sosm ---------------------------------------------------------------------- --------------------- Merged HDR images ------------------------------ ---------------------------------------------------------------------- merged_hdr_images contain hdr images generated by merging a sequence of multi-exposures with 5 HDR deghosting methods + 1 control condition for which images were merged without any deghosting (using pfstools). NOTE: Since most deghosting algorithms are computationally expensive, and often fail to process higher resolution images the resolution of sequence multi-exposures was rescaled to 1953×1301. This resolution is half the resolution of 16-bit tiff images obtained from captured RAW images. The structure of merged_hdr_images is organized as follows: merged_hdr_images complex image_set1 ghosted hu_setup1_complex.hdr nodegh_setup1_complex.hdr photomatix_setup1_complex.hdr photoshop_setup1_complex.hdr sen_setup1_complex.hdr silk_setup1_complex.hdr ground_truth hu_setup1_complex.hdr nodegh_setup1_complex.hdr photomatix_setup1_complex.hdr photoshop_setup1_complex.hdr sen_setup1_complex.hdr silk_setup1_complex.hdr image_set2 ghosted hu_setup2_complex.hdr nodegh_setup2_complex.hdr ..... ground_truth hu_setup2_complex.hdr nodegh_setup2_complex.hdr ..... image_set3 ..... image_set4 handheld .... sosm ---------------------------------------------------------------------- --------------- Subjective quality assessment data ------------------- ---------------------------------------------------------------------- The following files can be found in pairwise_comparison_results.zip: subjects_inputs.csv file contains the results of the pairwise comparison experiments. 'session_id' column indicates the session number (experiments were divided into 4 sessions due to lenghty time of experiments, each observer paraticipated in every session.) 'pair_number', the total number of comparison pairs per session is 150, thus pair number corresponds to the ordered pair number within each session. 'observer' referes to the anonymous observer name who participated in the experiment. 'scene' column indicates the name of the multi-exposure scene sequence which was processed with each deghosting algorithm and presented in the pairwise experiment. 'condition_1' and 'condition_2' corresponds to the pair of tonemapped images deghosted with a particular deghosting method that were presented to the observer side by side. 'selection' refers to the preferred 'condition' (i.e. deghosted image) selected by the observer. ---------------------------------------------------------------------- If you have any questions, please contact me via email (kanita.hadziabdic |.a.t.| gmail dot com). If you have any questions, please contact me by e-mail: (kanita.hadziabdic |.a.t.| gmail dot com).