Inception v3 different than Pytorch implementation

[m-parchami]

Hi,
hope you're doing well.
Is there any specific reason for the Inception v3 model (let's say for the CAM method as InceptionCAM), to be significantly different than the official Pytorch implementation? For starters, the Inception E and D blocks are completely removed. I checked the paper and couldn't find any references to this.

I would appreciate it if anyone can elaborate on this.
All the best.