tenggyut
tenggyut
Have tried that already...no luck. I ended up using tf.dropout instead of tl.layers.DropoutLayer...
> > Have tried that already...no luck. I ended up using tf.dropout instead of tl.layers.DropoutLayer... > > I tried doing the same, but ended up with a lot of other...
> @tenggyut I use tf.dropout instead of tl.layers.DropoutLayer as > > ```python > net.outputs = tf.nn.dropout(net.outputs, keep_prob=keep_rate, name='E_Dropout') > net_shape = net.outputs.get_shape() > ``` > > but get an error...
``` if data["imageData"] is not None: imageData = base64.b64decode(data["imageData"]) if PY2 and QT4: imageData = utils.img_data_to_png_data(imageData) else: # relative path from label file to relative path from cwd imagePath =...
I dont think the imagePath in json file is very much important, because the dataset may move around....
@Shahnawazgrewal based on our evaluation, your model is truly powerful than both provided pretrain-model. I wonder the reason behind this wonderful improvement? Is the dataset used for training the root...
@Shahnawazgrewal did you train the model as a classifier or using triple loss?
How to combine onet with faceboxes？use faceboxes's prediction as onet's input?
But the feature map generated by faceboxes is not reused, so may hurt the runtime efficiency? Also, did you reproduce the performance described in the original paper?
That's VERY KIND of you!