search

layumi / Image-Text-Embedding

TOMM2020 Dual-Path Convolutional Image-Text Embedding :feet: https://arxiv.org/abs/1711.05535
MIT License
285 stars 73 forks source link

How to set the learningrate when training CUHK-PEDES? #4

Closed jingyaa closed 6 years ago

jingyaa commented 6 years ago

Hi, I want to train the CUHK-PEDES using your method, in your paper "In the first training stage, we fixed the pre-trained image CNN, and train the text CNN only. The learning rate is 0.001.", https://github.com/layumi/Image-Text-Embedding/blob/master/train_cuhk_word2_1_pool.m#L34 and https://github.com/layumi/Image-Text-Embedding/blob/master/cuhk_word2_pool.m#L269 the learning rate is 0.1 and 0.001, how should I set the learning rate?

layumi commented 6 years ago

Hi @jingyaa Sorry for the late response. In matconvnet, the learning rate = global learning rate* layer learning rate. You may find the default layer learning rate at https://github.com/layumi/Image-Text-Embedding/blob/master/matlab/%2Bdagnn/%40DagNN/initParams.m

For text CNN, the learning rate of the final conv layer is 0.1\0.01 = 0.001, the other conv layer 0.1\0.1=0.01. For simplify, I just wrote the final learning rate 0.001 I will check again and make it clear in the revised version. Thank you!