search

MhLiao / TextBoxes_plusplus

TextBoxes++: A Single-Shot Oriented Scene Text Detector
Other
956 stars 281 forks source link

training parameters mismatch #57

Open freshcoderman opened 6 years ago

freshcoderman commented 6 years ago

I tried to train Textboxes_plusplus with my own dataset based on the pre-trained model on SynthText, but encountered the following error.

Cannot copy param 0 weights from layer 'conv4_3_norm_mbox_loc'; shape mismatch. Source param shape is 240 512 3 5 (1843200); target param shape is 144 512 3 5 (1105920). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer.

can you tell me which file should I modify so that they match each other?

Lily-two commented 6 years ago

May I ask you how to transform your image to xml.file? Thanks a lot .

YefeiGao commented 6 years ago
  1. generate xml format annotation file of ur annotation data, with pascal-voc format
  2. create train.txt / test.txt files by concatenating image and xml file path
  3. generate lmdb file for training @Lily-two
Lily-two commented 6 years ago

Get it. Thank you. @YefeiGao

tianranwangcs commented 6 years ago

@YefeiGao Are there any requirements for the sequence four vertexes in xml file? Like in clockwise or counterclockwise?

zq1335030905 commented 6 years ago

hello, I encountered the same problem,can i ask you how to solve it ? Thanks a lot