Implementing AssembleNet: Searching for Multi-Stream Neural Connectivity in Video Architectures Explain using Pytorch (Work In Process)

Reference

This repo is not official repository.

• Paper Link
• Author: Michael S. Ryoo (Robotics at Google, Google Research), AJ Piergiovanni (Robotics at Google, Google Research), Mingxing Tan (Google Research), Anelia Angelova (Robotics at Google, Google Research)
• Organization: Robotics at Google, Google Research

Usage

• Make graph

from make_graph import Graph
import pprint

p = pprint.PrettyPrinter(width=160, indent=4)
g = Graph()
p.pprint(g.grpah)

• Make Network

from make_graph import Graph
import pprint

g = Graph()
m = Model(g.graph)
pprint.pprint(m.graph, width=160)

x = torch.randn([2, 3, 16, 256, 256])
print(m(x).size())

• Network Evolution

from make_graph import Graph
import pprint

g = Graph()
m = Model(g.graph)
pprint.pprint(m.graph, width=160)
m._evolution()
pprint.pprint(m.graph, width=160)

x = torch.randn([2, 3, 16, 256, 256])
print(m(x).size())

Evolution

Work In Process

• Connection-Learning-Guided Mutation
• Evolution
• Training