CharlesShang / TFFRCNN

FastER RCNN built on tensorflow
MIT License
874 stars 418 forks source link

The concat layer before rpn in pvanet #56

Open wew2012 opened 7 years ago

wew2012 commented 7 years ago
        (self.feed('conv5_4/last_relu')
         .upconv(tf.shape(self.layers['downsample']),
                 384, 4, 2, name = 'upsample', biased= False, relu=False, trainable=True)) # upsample

        (self.feed('downsample', 'conv4_4')
         .concat(axis=3, name='concat'))

In lib/networks/PVAnet_train.py:92, the concat layer before rpn concat 'downsample' and 'conv4_4' layer, which is different from the paper. In the paper, 'upsample' layer should be used too. I wonder if I made a mistake.