ProGamerGov

@htoyryla Ok, thanks for the help!

So I successfully create the lmdb files! https://gist.github.com/ProGamerGov/d0038f7e3186d057bb7b26398bd764f9 It seems that a few of the images listed in the train.txt and val.txt files, did not exist in the actual data...

Trying to start the fine tuning, seems to be throwing out an error: ``` ubuntu@ip-Address:~/caffe$ ./build/tools/caffe train -solver models/vgg16_finetune/solver.prototxt -weights models/vgg16_finetune/VGG16_SOD_finetune.caffemodel.caffemodel -gpu 0 libdc1394 error: Failed to initialize libdc1394 I0728...

So everything was working well until this happened: `Memory required for data: 1152053324`

``` iter_300 accuracy = 0.25 iter_400 accuracy = 0.31 iter_600 accuracy = 0.21 iter_700 accuracy = 0.22 iter_800 accuracy = 0.16 iter_900 accuracy = 0.23 iter_1300 accuracy = 0.21 iter_1500...

Test 9: ``` batch_size: 12 batch_size: 5 base_lr: 0.0005 ``` Accuracy: | Iterations | Accuracy | | :-- | :-- | | 100 | accuracy = 0 | | 200...

@htoyryla Thanks, I missed that mistake. Hopefully that will help with the accuracy value. Though I may have to play around more with the `base_lr` and `batch_size` values because I...

This is the prob layer and I just comment it out like this, correct? ``` #layers { # bottom: "fc8_43" # top: "prob" # name: "prob" # type: SOFTMAX #}...

Should I be using the "type" settings in the solver.prototxt for fine tuning? If so which of the 6 options should I be using? ``` 1. Stochastic Gradient Descent "SGD"...

My current test that starts at iteration 800 with a test every 100 iterations: ``` accuracy = 0.25 accuracy = 0.275 accuracy = 0.15 accuracy = 0.2 accuracy = 0.2875...