资源分类
》 资源 》算法》GoogLeNet GPU implementation from Princeton.

GoogLeNet GPU implementation from Princeton.

2019-09-20 | |  122 |   0 |   0

We implemented GoogLeNet using a single GPU. Our main contribution is an effective way to initialize the network and a trick to overcome the GPU memory constraint by accumulating gradients over two training iterations.
* Please check http://3dvision.princeton.edu/pvt/GoogLeNet/ for more information. Pre-trained models on ImageNet and Places, and the training code are available for download.
* Make sure cls2_fc2 and cls3_fc have num_output = 1000 in the prototxt. Otherwise, the trained model would crash on test.

无链接

上一篇：Places-CNN model from MIT.

下一篇：CaffeNet fine-tuned for Oxford flowers dataset

用户评价
全部评价

热门资源

  • DuReader_QANet_BiDAF

    Machine Reading Comprehension on DuReader Usin...

  • tensorflow-sketch...

    Discrlaimer: This is not an official Google pro...

  • My_DrQA

    My_DrQA A re-implement DrQA based on Pytorch

  • ETD_cataloguing_a...

    ETD catalouging project using allennlp

  • allennlp_extras

    allennlp_extras Some utilities build on top of...