search

CSAILVision / NetDissect

Network Dissection http://netdissect.csail.mit.edu for quantifying interpretability of deep CNNs.
MIT License
449 stars 115 forks source link

Reproducing results #24

Open KrishnaTarun opened 2 years ago

KrishnaTarun commented 2 years ago

I'm trying to run resnet152 trained on ImageNet using the python wrapper provided in the repository with the standard settings but the results are not the same as in the paper. Any suggestions for that.

KrishnaTarun commented 2 years ago

Attached is the result for Resnet152 (Top) and Alexnet (Bottom) trained on ImageNet and Places 365 respectively. Both the models seem to detect only scenes and textures which is not the case if we observe Figure 7 in https://netdissect.csail.mit.edu/final-network-dissection.pdf. Selection_029 Selection_036

Any comments on this would be helpful. Thanks.