What is the "fast version" of ZFNet referenced in SPPNet and Faster R-CNN papers?

Question

I'm reading old papers:

• SPPNet: Link
• Faster R-CNN: Link

In both cases, the authors refer to a "fast version of Zeiler and Fergus (ZF) Net"; specifically:

• In SPPNet:

  ZF-5: this architecture is based on Zeiler and Fergus’s (ZF) “fast” (smaller) model [4]. The number indicates five convolutional layers.

• In Faster R-CNN:

  we use the “fast” version of ZF net [32] that has five convolutional layers and three fully-connected layers..

where both [4] and [32] refer to the same famous paper Visualizing and Understanding Convolutional Networks by D. Zeiler and R. Fergus.

My question: what is the "fast" version of ZFNet?

Searching for ZFNet and reading the paper, there seems to be only one architecture with five convolutional layers and three fully-connected layers. What exactly differentiates the "fast" version from any other version of ZFNet mentioned in subsequent papers?

Answer 1

The "fast" version uses all convolutional layers and fully connected layers in comparison to the other versions, in which they remove several different layers. This version is likely called the "fast" version because it is can predict results faster as a result of having more convolutional layers. This downsamples the image further before getting to the fully connected layers, which leads to a lower number of paramaters in the model.