maybeshewill-cv / nsfw-classification-tensorflow Goto Github PK

I am trying to build my own model, however I am having difficulty actually starting on that. Your GitHub has been quite helpful, however I was trying your version vs the one that nsfwjs.com uses under the 'Inception V3 model' and I noticed that despite the same dataset used in both yours and their model, your version has significant false positives. I am trying to understand how that can be? and if you're using a different base model that you built off of or something else?

for example:
https://i.imgur.com/w5FPQbn.png vs https://i.imgur.com/rKvuy0U.png and https://i.imgur.com/cnzGpQW.png

I tried substituting their model (https://github.com/infinitered/nsfwjs/tree/master/example/nsfw_demo/public/model) with your model (https://github.com/MaybeShewill-CV/MaybeShewill-CV.github.io/tree/master/nsfw_classification/model) and it didn't load, so I aim imagining that there was a significant difference in process from your method and theres?

thank you for making this github though!! really helpful

Hello:
when I used the test_nsfw.py to predict image with the model/nsfw_export_saved_model/1/variables/variables.index

I encountered the following problem:
W tensorflow/core/framework/op_kernel.cc:1192] Not found: Key nsfw_cls_model/resnet_block_layer_3/block_fn_4/bn_1/beta/ExponentialMovingAverage not found in checkpoint
which means when training the saver errors,it is wrong by saver = tf.train.Saver() to store the models