SilenceMonk-CSDN博客

原创 score-based generative models (Yang Song) 笔记

20220524Yang Song’s blog文章目录20220524方法方法The score function, score-based models, and score matchingFisher divergencescore：从当前 xxx 指向 p(x)p(x)p(x) 更大的方向联系 diffusion model：对当前 xxx 的噪声估计如何直接估计 gt score ？同 flow 讲，对 SθS_{\theta}Sθ 没有约束Langevi

2022-05-24 23:45:17 2279

原创 How Distance Transform Maps Boost Segmentation CNNs: An Empirical Study

20220523JMLR2020主要讲了下distance transform在医学图像分割的应用方法模型预测 distance map 的方法：使用轻量级的 head 或者单独的 decodersdf两大类利用方式：loss/auxiliary tasklossBoundary Loss知乎：JunMa实质上是使用 sdf 对普通 BCE 加权？：sdf 不是有符号？如果prediction和label一致，loss为0。如果prediction比label小并被

2022-05-23 14:42:02 419

原创作业一：推导 Softmax 梯度

作业一：推导 Softmax 梯度Created: March 18, 2022 1:19 PML=−log⁡esk∑jesjL = - \log \frac{e^{s_k}}{\sum_j e^{s_j}}L=−log∑jesjesk,求 ∂L∂esi\frac{\partial L}{\partial e^{s_i}}∂esi∂L 。定义 pi=esi∑jesjp_i = \frac{e^{s_i}}{\sum_j e^{s_j}}pi=∑jesjesii=ki =

2022-03-25 16:44:28 788 1

原创 pytorch k 折

### Load datasettrainset = datasets.ImageFolder(train_dir, transform=transform)testset = datasets.ImageFolder(test_dir, transform=transform)import torchfrom sklearn.model_selection import KFolddata_induce = np.arange(0, len(trainset)) # 将“训练集”分为训练

2022-03-25 00:45:25 1969

原创 matplotlib colormap

原文章原文章排版有问题，我整理了下 get_cm，记录在这def get_jet_cm(name='jet'): colormap_int = np.zeros((256, 3), np.uint8) colormap_float = np.zeros((256, 3), np.float) for i in range(0, 256, 1): colormap_float[i, 0] = cm.__dict__[name](i)[0] colo

2022-03-23 15:11:04 2812

原创 Transferable Semi-Supervised Semantic Segmentation

AAAI 2018National University of Singapore

2021-09-23 00:35:03 63

原创 Weakly-Supervised Semantic Segmentation Network with Deep Seeded Region Growing

awesome-semi-supervised-learningWeakly-Supervised Semantic Segmentation Network withDeep Seeded Region GrowingCVPR 2018MSRA

2021-09-18 10:41:13 158

原创 Revisiting Dilated Convolution:A Simple Approach for WeaklyAnd SemiSupervised Semantic Segmentation

Revisiting Dilated Convolution:A Simple Approach for WeaklyAnd SemiSupervised Semantic SegmentationCVPR 2018，weakly supervised segmentationUIUC知乎笔记

2021-09-17 23:28:07 226

原创 PyTorch Mobile FloatArray 转 Bitmap （Android/Kotlin)

'org.pytorch:pytorch_android:1.4.0'的TensorImageUtils.imageYUV420CenterCropToFloat32Tensor()是真的天坑，进来的图像RGB是扭曲的（目前），希望 PyTorch Mobile 以后会改。。用 PyTorch Mobile 最后要生成图片之前必须要用TensorImageUtils.bitmapToFloat32Tensor()至于读进来yuv格式图片的同学们要自己想办法yuv转bit.

2020-12-03 00:50:59 992 2

m0_48064374的博客