Search Results for author:
Found 11 papers, 3 papers with code
Scale Equalization for Multi-Level Feature Fusion
We discover that scale disequilibrium is caused by bilinear upsampling, which is supported by both theoretical and empirical evidence.
Analysis of NaN Divergence in Training Monocular Depth Estimation Model
The latest advances in deep learning have facilitated the development of highly accurate monocular depth estimation models.
Resolution-Aware Design of Atrous Rates for Semantic Segmentation Networks
However, fixed values of atrous rates are used for the ASPP module, which restricts the size of its field of view.
Ranked #1 on
Retinal Vessel Segmentation
on HRF
(mIoU metric)
Understanding Gaussian Attention Bias of Vision Transformers Using Effective Receptive Fields
Based on this observation, we propose explicitly adding a Gaussian attention bias that guides the positional embedding to have the corresponding pattern from the beginning of training.
How to Use Dropout Correctly on Residual Networks with Batch Normalization
In this study, we investigate the correct position to apply dropout.
On the Ideal Number of Groups for Isometric Gradient Propagation
First, we find that the number of groups influences the gradient behavior of the group normalization layer.
Guidelines for the Regularization of Gammas in Batch Normalization for Deep Residual Networks
L2 regularization for weights in neural networks is widely used as a standard training trick.
Ranked #2 on
Text Classification
on GLUE SST2
Improved Robustness of Vision Transformer via PreLayerNorm in Patch Embedding
We compared the robustness of CNN and ViT by assuming various image corruptions that may appear in practical vision tasks.
Dead Pixel Test Using Effective Receptive Field
First, we evaluated the size of the receptive field.
Ranked #5 on
Fine-Grained Image Classification
on Caltech-101
Extending Class Activation Mapping Using Gaussian Receptive Field
First, we propose Gaussian upsampling, an improved upsampling method that can reflect the characteristics of deep learning models.
Selective Distillation of Weakly Annotated GTD for Vision-based Slab Identification System
The main contribution of this paper is on selective distillation to improve the quality of the weakly annotated GTD.
