Anson Yu
Anson Yu
Hi, Sorry, the script is indeed kind of messy! I think you can try modifying line 51 to `exit(0)` instead of `break`
Try adding `print(number_of_pictures)` on next line of line 14, what does it print?
Yes, this should work!
Hi Hong, I think the problem is wrong indentation compared to the code microhua pasted. The code with the line starting from read_img_name should be indented to the right. ```...
Hi, I think it doesn't affect the results a lot, I just chose 3000 scenery pictures from the [IMAGENET](http://image-net.org/) dataset.
Hi, Please refer to issue #25 . Just keep adjusting the thresholds and confidence further if you still see a lot of false positives. Thanks, Anson
Maybe you can try using **face_cascade_fullconv_fddb_single_crop.py** to do the detection, since it takes 1 crop instead of 10 when feedforwarding images. I think the avg. time using GTX 970 was...
I believe this can be done by the following steps: 1. Add the 12-net layers to the 24-net .prototxt files (concatenating the fully connected layers as shown in the paper)...
If during training, the accuracy on the validation set is higher than 0.9, 12-net should perform great during testing. However, if the accuracy stays fixed at 0.5, you may need...
Hi, I think you can simply create the lmdb files as before, and for example in face_24's [train_val](https://github.com/anson0910/CNN_face_detection_models/blob/master/face_24c/train_val.prototxt) file, add a data layer specifying the **data_param**.