SSD300MobileNet predicts bad results

tanakataiki / ssd_kerasV2

BetterNetWork4SSD

MIT License

144 stars 56 forks source link

SSD300MobileNet predicts bad results #17

Open jescur opened 5 years ago

jescur commented 5 years ago

Hi there,

I tried to run the code in SSD.ipynb just like you have it: with the model ssd300MobileNet, and picking the weights you provide at https://drive.google.com/drive/u/0/folders/1F8GjD3BFhf_hv9Ipez0twRptYc3P8YwP (MobileNetSSD300weights_voc_2007.hdf5), and I get these results:

cat_car kids

you can see they have nothing to do with the results you present.

Could you tell me what can be wrong? Are the weights the correct ones? Is the model working?

Thanks!

tanakataiki commented 5 years ago

It's not fully trained yet so you need to retrain father. but using pretrained weight is at least better than trainning from scratch

luismikg commented 5 years ago

Hi! I have got the same issue but I think I have found the solution: on this line: model.load_weights('./MobileNetSSD300weights_voc_2007.hdf5', by_name=True) change by: model.load_weights('./MobileNetSSD300weights_voc_2007.hdf5', by_name=False) and try again. I cannot understand what it is the meaning of the parameter "by_name", Please tell us if this solution works for you in order to help all users to understand better this model.

luismikg commented 5 years ago

Ok!, I have reviewed this model again, I have discovered that the model has changed and now to set by_name=False get an error. If you retrain it. The model will work better. However, It means that the first weights Maybe do they have a kind of error?