Most convolutional neural networks use some method for gradually downscaling the size of the hidden layers. This is commonly referred to as pooling, and is applied to reduce the number of parameters, improve invariance to certain distortions, and increase the receptive field size. Since pooling by nature is a lossy process, it is crucial that each such layer maintains the portion of the activations that is most important for the network's discriminability. Yet, simple maximization or averaging over blocks, max or average pooling, or plain downsampling in the form of strided convolutions are the standard. In this paper, we aim to leverage recent results on image downscaling for the purposes of deep learning. Inspired by the human visual system, which focuses on local spatial changes, we propose detailpreserving pooling (DPP), an adaptive pooling method that magnifies spatial changes and preserves important structural detail. Importantly, its parameters can be learned jointly with the rest of the network. We analyze some of its theoretical properties and show its empirical benefits on several datasets and networks, where DPP consistently outperforms previous pooling approaches.

No abstract

Kopf et al. [2013]Ö ztireli and Gross [2015] DPID λ=1.0 DPID λ=0.5 Figure 1: Row 1: Input images with 0.5, 1.9, 2.7, and 4.6 megapixels respectively. Rows 2-5: Downscaled results with 128 pixels width. Our algorithm (DPID) preserves stars in Example 1, thin lines in Example 2, roof tiles in Example 3, and text, lines and notes in Example 4.

This study is a first synthesis focused on incised-valleys located within the inner shelf of the Bay of Biscay. It is based on previously published results obtained during recent seismic surveys and coring campaigns. The morphology of the valleys appears to strongly controlled by tectonics and lithology. The Pleistocene sedimentary cover of the shelf is very thin and discontinuous with a maximum thickness ranging between 30 and 40 m in incised valley fills. Thus the incised bedrock morphology plays a key-role by controlling hydrodynamics and related sediment transport and deposition that explains some variations of those incisedvalley fills with respect to the previously published general models.Present-day seismic activity in the western part of France is moderate (Müller et al., 1992). It is possibly related to strain originating at the southern European plate boundary (Ziegler, 1992).Onland, rocky outcrops correspond to two main sets: (1) metamorphic and magmatic rocks related to the Hercynian orogen, north of the lay-Sèvre estuary (Pertuis breton, Fig. 1) and (2) Mesozoic and Cenozoic sedimentary rocks including, from north to south, early Jurassic to Pliocene strata in the Aquitanian basin, south of the Lay-Sèvre estuary. Southward of the Gironde estuary, Cenozoic strata are overlain by an extensive Quaternary sand cover including large eolian sand dunes.

No abstract