chr5tphr / zennit

Zennit is a high-level framework in Python using PyTorch for explaining/exploring neural networks using attribution methods like LRP.

chr5tphr/zennit Issues

Unexpected behavior with object.__setattr__ in canonizers.MergeBatchNorm
Updated a month ago
Potential issue with learning BatchNorm parameters
Updated 9 months ago
ResNet: Unstable Attributions
Updated 9 months ago3
Model inference outside of composite context
Updated 9 months ago5
Changing The Value of epsilon in the composite Epsilonplusflat creates a bug
Closed 9 months ago2
SmoothGrad - potential typo in implementation of noise level
Closed 10 months ago2
Neuralized K-Means: make k-means amenable to neural network explanations
Updated a year ago
Proper way of handling classifier heads in Transformers
Updated a year ago2
Core/Composite: Simplify Composite.context
Updated a year ago
Module with Multiple Inputs
Updated a year ago2
Core: GradOutHooks
Updated a year ago
Core: Warn the user when falling back to gradient
Updated a year ago
Numerical instability in ResNet50 heatmaps
Updated a year ago12
Custom rules for vision transformer
Updated a year ago1
LinearAttention Module
Updated a year ago1
How to add support to DenseNet from torchvision
Updated a year ago2
Layer-wise LRP score
Updated a year ago6
Canonizers: BatchNorm loses bias relevance
Updated a year ago
Smooth MaxPool2D rule
Updated a year ago2
Docs: Build Tutorials on ReadTheDocs with trained weights
Updated a year ago
support unet with Upsample or ConvTranspose2d layer?
Closed a year ago29
Obtanining graident of LRP otput w.r.t. network parameters
Closed a year ago12
Feature Request: Debugger
Updated a year ago2
Bug: MultiheadAttention in types
Closed a year ago4
Binary classification model with one output for output layer
Closed a year ago1
Composite: bug in BetaSmooth constructor?
Closed 2 years ago3
error occurred when run tutorial on custom model
Closed 2 years ago7
Core: Second Order Gradients
Closed 2 years ago4
Not removing backward hook handler *might* cause problems?
Closed 2 years ago1
Support for padding layers
Closed 2 years ago2
Notebook for feed_forward.py
Closed 2 years ago8
Bug in AlphaBeta rule?
Closed 3 years ago2
Feature request: Assign rule by layer index
Closed 3 years ago2
Feature request: Support for max-pooling layers
Closed 3 years ago2
Canonizers
Closed 3 years ago2
torch 1.10 version not working
Closed 3 years ago2
RuntimeError: Mismatch in shape: grad_output[0] has a shape of torch.Size([1, 1, 80, 80]) and output[0] has a shape of torch.Size([1, 1, 96, 96]).
Closed 3 years ago
RuntimeError: Mismatch in shape: grad_output[0] has a shape of torch.Size([1, 1, 80, 80]) and output[0] has a shape of torch.Size([1, 1, 96, 96]).
Closed 3 years ago
Alpha Beta still includes Bias two times
Closed 3 years ago1
Default Rule
Closed 3 years ago2
ResNet canonizer misses canonizer for BasicBlock module
Closed 3 years ago
New version of zennit produces wrong LRP heatmaps
Closed 3 years ago4
[How to use] heatmap variants [...] such as the guided_backprop or smoothgrad [...] ?
Closed 3 years ago2
feed_forward is broken for torchvision 0.10.0
Closed 3 years ago
Switch over to Module.register_full_backward_hook
Closed 3 years ago4
Image fetching in example is broken
Closed 3 years ago
Flat Rule for Pooling Layers
Closed 3 years ago4
Memory Leak
Closed 3 years ago2