Images or videos always contain multiple objects or actions. Multi-label recognition has been witnessed to achieve pretty performance attribute to the rapid development of deep learning technologies. Recently, graph convolution network (GCN) is leveraged to boost the performance of multi-label recognition. However, what is the best way for label correlation modeling and how feature learning can be improved with label system awareness are still unclear. In this paper, we propose a label graph superimposing framework to improve the conventional GCN+CNN framework developed for multi-label recognition in the following two aspects. Firstly, we model the label correlations by superimposing label graph built from statistical co-occurrence information into the graph constructed from knowledge priors of labels, and then multi-layer graph convolutions are applied on the final superimposed graph for label embedding abstraction. Secondly, we propose to leverage embedding of the whole label system for better representation learning. In detail, lateral connections between GCN and CNN are added at shallow, middle and deep layers to inject information of label system into backbone CNN for label-awareness in the feature learning process. Extensive experiments are carried out on MS-COCO and Charades datasets, showing that our proposed solution can greatly improve the recognition performance and achieves new state-of-the-art recognition performance.

No abstract

features for developing high-performance X-ray imagers and ultraviolet sensors. [3] For imaging application, IGZO TFTs were mainly integrated with amorphous silicon and organic photodiodes, [3c,e,4] the spectral of which were usually limited to visible light detection. Large-area, active-matrix infrared sensing, therefore, remains a challenge.Organic-inorganic hybrid perovskites possess excellent optoelectronic properties with high absorption coefficient, high charge carrier mobility, long carrier diffusion lengths, and tunable bandgaps, [5] rendering it a promising material for across-the-board optoelectronic devices, [6] including photovoltaic cells, [7] light-emitting, lasing devices, [8] and high-performance photodetectors. [9] In particular, compared with predominant photodetectors made of inorganic semiconductors, solution-processable perovskite is more promising for lowcost, flexible, and large-area scenarios. [6d] Conventional lead-based perovskite photodiodes (PDs) provide a detection spectral range from 300 to 800 nm. [10] To further extend the spectral response to the near-infrared (NIR) range, there are normally two approaches. One is to combine perovskite with narrow bandgap polymers or quantum dots such as PDPP3T [11] and PbS quantum dots. [12] Another approach is to introduce an Sn-Pb binary perovskite PDs. [13] The smaller ionic radii of Sn 2+ than Pb 2+ (Sn 2+ :1.35 Å and Pb 2+ : 1.49 Å) [14] reduces the bandgap of perovskites due to the bowing effect, [15] so that Sn-Pb hybrid perovskite has lower bandgap to extend the light absorption to ≈1000 nm. Recently Xiaobao et al. reported MA 0.5 FA 0.5 Pb 0.5 Sn 0.5 I 3 perovskite photodetector, exhibiting a detectivity of over 10 12 Jones ranging from 800 to 970 nm. [13c] Wang et al. fabricated mixed Sn-Pb perovskite photodetectors with a broadband response from 300 to 1000 nm, responsivity (R) of over 0.4 A W −1 , and detectivity (D*) of over 10 12 Jones in the near-infrared region. [13b] Encouraging improvements in the Sn-Pb based perovskite stability have also been reported very recently, such as using GuaSCN passivation to achieve 1 µs carrier lifetime with long-term stability. [16] The above development of low bandgap organic-inorganic perovskite materials has brought new opportunity for developing advanced flat-panel Flat-panel imagers have wide applications in industrial and medical inspections. Nonetheless, large area infrared imaging remains a challenge due to the fact that the state-of-the-art infrared sensors are usually based on silicon or germanium technologies, which are limited by the wafer size. Recent advances in low bandgap Sn-Pb perovskite photodiodes (PDs) and indium gallium zinc oxide (IGZO) thin-film transistors (TFTs) matrix backplane bring new opportunity for developing the large area near-infrared image sensor. As a proof of concept, a 12 × 12 pixels array with each pixel independently controlled by the gate voltage of a TFT are constructed. Arrays of Sn-Pb based perovskite PDs are spin deposited onto the IGZO TF...