Autocropping: A Closer Look at Benchmark Datasets

Celona, Luigi; Ciocca, Gianluigi; Napoletano, Paolo; Schettini, Raimondo

doi:10.1007/978-3-030-30645-8_29

Cited by 1 publication

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We consider other common aspect ratios in addition to those mentioned above for this visualization. We also compute the mean absolute error of each image to the closest aspect ratio in Figure 1c, similar to the analysis performed by Celona et al [2].…”

Section: A Datasetmentioning

confidence: 99%

“…We use the Boundary Displacement Error (BDE) and the Intersection over Union (IoU) to evaluate cropping, in line with previous approaches [2], [13], [22], [24], [32]. As we did not have editors rank or rate different crops, we cannot compute ranking metrics or leverage ranked learning approaches.…”

Section: Experiments and Ablation Studymentioning

confidence: 99%

“…We compare our models with a baseline method that predicts bounding boxes of varying sizes in the correct aspect ratio around the center of the image [2]. We denote this family of methods, Baseline-s, where s represents the scaling factor of the bounding box as a fraction of the largest possible bounding box for that aspect ratio.…”

Section: A Evaluation On Our Datasetmentioning

confidence: 99%

See 2 more Smart Citations

An Experience-Based Direct Generation Approach to Automatic Image Cropping

Christensen

Vartakavi

2021

IEEE Access

View full text Add to dashboard Cite

Automatic Image Cropping is a challenging task with many practical downstream applications. The task is often divided into sub-problems -generating cropping candidates, finding the visually important regions, and determining aesthetics to select the most appealing candidate. Prior approaches model one or more of these sub-problems separately, and often combine them sequentially. We propose a novel convolutional neural network (CNN) based method to crop images directly, without explicitly modeling image aesthetics, evaluating multiple crop candidates, or detecting visually salient regions. Our model is trained on a large dataset of images cropped by experienced editors and can simultaneously predict bounding boxes for multiple fixed aspect ratios. We consider the aspect ratio of the cropped image to be a critical factor that influences aesthetics. Prior approaches for automatic image cropping, did not enforce the aspect ratio of the outputs, likely due to a lack of datasets for this task. We, therefore, benchmark our method on public datasets for two related tasks -first, aesthetic image cropping without regard to aspect ratio, and second, thumbnail generation that requires fixed aspect ratio outputs, but where aesthetics are not crucial. We show that our strategy is competitive with or performs better than existing methods in both these tasks. Furthermore, our one-stage model is easier to train and significantly faster than existing two-stage or endto-end methods for inference. We present a qualitative evaluation study, and find that our model is able to generalize to diverse images from unseen datasets and often retains compositional properties of the original images after cropping. We also find that the model can generate crops with better aesthetics than the ground truth in the MIRThumb dataset for image thumbnail generation with no fine tuning. Our results demonstrate that explicitly modeling image aesthetics or visual attention regions is not necessarily required to build a competitive image cropping algorithm.

show abstract

Section: A Datasetmentioning

confidence: 99%