About Regression + Classification and Pretraining

[HEUzhouhanwen]

Hi！
In paper 'Real-Time Grasp Detection Using Convolutional Neural Networks'
part IV. GRASP DETECTION WITH NEURAL NETWORKS
C. Regression + Classification
D. MultiGrasp Detection
Do you have a program?
part V. EXPERIMENTS AND EVALUATION
C. Pretraining
Is models/imagenet/m2/m2.ckpt the weight of pre-training results saved？
or models/imagenet/m4/m4.ckpt
Thank you！
Shu Guo！

[HEUzhouhanwen]

The loss is stable at 30000-40000，I think it is related to Pretraining！

[tnikolla]

I don't have a program for Reg+Class or MultiGrasp. Fill free to contribute.

models/imagenet/... are the saved parameters for imagenet.
models/grasp/... are the saved parameters for grasping dataset.

[zhoumo1121]

ValueError: Tensor conversion requested dtype int32 for Tensor with dtype float64: 'Tensor("truediv:0", shape=(), dtype=float64, device=/device:CPU:0)'
May I ask how to solve this problem? Thanks, I've been working on it for a long time!