干货|PyTorch实用代码段集锦

Facebook公司开源的深度学习框架PyTorch越来越火，PyTorch易于上手。本文节选github中PyTorch的常用实用代码片段，供大家学习参考。

Github链接地址：

https://github.com/ptrblck/pytorch_misc

代码总览

• accumulate_gradients - Comparison of accumulated gradients/losses to vanilla batch update. 链接：https://github.com/ptrblck/pytorch_misc/blob/master/accumulate_gradients.py
• adaptive_batchnorm- Adaptive BN implementation using two additional parameters: out = a * x + b * bn(x). 链接： https://github.com/ptrblck/pytorch_misc/blob/master/adaptive_batchnorm.py
• adaptive_pooling_torchvision - Example of using adaptive pooling layers in pretrained models to use different spatial input shapes. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/adaptive_pooling_torchvision.py
• batch_norm_manual - Comparison of PyTorch BatchNorm layers and a manual calculation. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/batch_norm_manual.py
• change_crop_in_dataset - Change the image crop size on the fly using a Dataset. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/change_crop_in_dataset.py
• conv_rnn - Combines a 3DCNN with an RNN; uses windowed frames as inputs. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/conv_rnn.py
• csv_chunk_read - Provide data chunks from continuous .csv file. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/csv_chunk_read.py
• densenet_forwardhook - Use forward hooks to get intermediate activations from densenet121. Uses separate modules to process these activations further. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/densenet_forwardhook.py
• edge_weighting_segmentation - Apply weighting to edges for a segmentation task. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/edge_weighting_segmentation.py
• image_rotation_with_matrix - Rotate an image given an angle using 1.) a nested loop and 2.) a rotation matrix and mesh grid. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/image_rotation_with_matrix.py
• LocallyConnected2d - Implementation of a locally connected 2d layer. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/LocallyConnected2d.py
• mnist_autoencoder - Simple autoencoder for MNIST data. Includes visualizations of output images, intermediate activations and conv kernels. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/mnist_autoencoder.py
• mnist_permuted - MNIST training using permuted pixel locations. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/mnist_permuted.py
• model_sharding_data_parallel - Model sharding with DataParallel using 2 pairs of 2 GPUs. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/model_sharding_data_parallel.py
• momentum_update_nograd - Script to see how parameters are updated when an optimizer is used with momentum/running estimates, even if gradients are zero. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/momentum_update_nograd.py
• shared_array - Script to demonstrate the usage of shared arrays using multiple workers. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/shared_array.py
• unet_demo - Simple UNet demo. 链接： https://github.com/ptrblck/pytorch_misc/blob/master/unet_demo.py
• weighted_sampling - Usage of WeightedRandomSampler using an imbalanced dataset with class imbalance 99 to 1 链接： https://github.com/ptrblck/pytorch_misc/blob/master/weighted_sampling.py