Multiscale features are of great importance in modern convolutional neural networks, showing consistent performance gains on numerous vision tasks. Therefore, many plug-and-play blocks are introduced to upgrade existing convolutional neural networks for stronger multiscale representation ability. However, the design of plug-and-play blocks is getting more and more complex, and these manually designed blocks are not optimal. In this work, we propose PP-NAS to develop plug-and-play blocks based on neural architecture search (NAS). Specifically, we design a new search space PPConv and develop a search algorithm consisting of one-level optimization, zero-one loss, and connection existence loss. PP-NAS minimizes the optimization gap between super-net and subarchitectures and can achieve good performance even without retraining. Extensive experiments on image classification, object detection, and semantic segmentation verify the superiority of PP-NAS over state-of-the-art CNNs (e.g., ResNet, ResNeXt, and Res2Net). Our code is available at https://github.com/ainieli/PP-NAS.

Automatic data augmentation is a technique to automatically search for strategies for image transformations, which can improve the performance of different vision tasks. RandAugment (RA), one of the most widely used automatic data augmentations, achieves great success in different scales of models and datasets. However, RA randomly selects transformations with equivalent probabilities and applies a single magnitude for all transformations, which is suboptimal for different models and datasets. In this paper, we develop Differentiable RandAugment (DRA) to learn selecting weights and magnitudes of transformations for RA. The magnitude of each transformation is modeled following a normal distribution with both learnable mean and standard deviation. We also introduce the gradient of transformations to reduce the bias in gradient estimation and KL divergence as part of the loss to reduce the optimization gap. Experiments on CIFAR-10/100 and ImageNet demonstrate the efficiency and effectiveness of DRA. Searching for only 0.95 GPU hours on ImageNet, DRA can reach a Top-1 accuracy of 78.19% with ResNet-50, which outperforms RA by 0.28% under the same settings. Transfer learning on object detection also demonstrates the power of DRA. The proposed DRA is one of the few that surpasses RA on ImageNet and has great potential to be integrated into modern training pipelines to achieve state-of-the-art performance. Our code will be made publicly available for out-of-the-box use.

Glioma grading during surgery can help clinical treatment planning and prognosis, but intraoperative pathological examination of frozen sections is limited by the long processing time and complex procedures. Near-infrared fluorescence imaging provides chances for fast and accurate real-time diagnosis. Recently, deep learning techniques have been actively explored for medical image analysis and disease diagnosis. However, issues of near-infrared fluorescence images, including small-scale, noise, and low-resolution, increase the difficulty of training a satisfying network. Multi-modal imaging can provide complementary information to boost model performance, but simultaneously designing a proper network and utilizing the information of multi-modal data is challenging. In this work, we propose a novel neural architecture search method DLS-DARTS to automatically search for network architectures to handle these issues. DLS-DARTS has two learnable stems for multi-modal low-level feature fusion and uses a modified perturbation-based derivation strategy to improve the performance on the area under the curve and accuracy. White light imaging and fluorescence imaging in the first near-infrared window (650-900 nm) and the second near-infrared window (1,000-1,700 nm) are applied to provide multi-modal information on glioma tissues. In the experiments on 1,115 surgical glioma specimens, DLS-DARTS achieved an area under the curve of 0.843 and an accuracy of 0.634, which outperformed manually designed convolutional neural networks including ResNet, PyramidNet, and EfficientNet, and a state-of-the-art neural architecture search method for multi-modal medical image classification. Our study demonstrates that DLS-DARTS has the potential to help neurosurgeons during surgery, showing high prospects in medical image analysis.