rockyzhengwu / nsfw Goto Github PK

上传了我儿子图片 结果竟然这样的

File "nsfw_predict.py", line 42, in predict
feed_dict={inputs: [image_data] * _BATCH_SIZE})
File "\nsfw-master\venv\lib\site-packages\tensorflow\python\client\session.py", line 950, in run
run_metadata_ptr)
File "\nsfw-master\venv\lib\site-packages\tensorflow\python\client\session.py", line 1149, in _run
str(subfeed_t.get_shape())))
ValueError: Cannot feed value of shape (128, 64, 64) for Tensor 'input_tensor:0', which has shape '(128, 64, 64, 3)'

I test the model with a few pitcures which is the close up of the sexual organs. The model fails to classify them and concludes that these pitcures are suitable for work.

使用http://ai.midday.me 鉴别yellow图功能检测出来的分数，和手动运行python nsfw_predict.py + 路径 得到的分数不一样，有时甚至差很多，以至于分类不同。
是模型不一样么？感觉线上的检测效果会更好。
如果模型不一样，能不能提供一下线上的模型呀？感谢！

When i try to convert this model to tflite , i get this error. any idea how to fix this ?

b'/bin/sh: toco_from_protos: command not found\n' None

Code: converter = lite.TFLiteConverter.from_saved_model(_MODEL_DIR) tflite_model = converter.convert()

finetune的是哪个模型呀……我试了resnet预训练好的几个模型，都结构对不上

Hi! I have trained a new model via running nsfw_main_finetune.py and got some ckpt files. But how may I use ckpt files to predict a test image without converting them to pb files? thanks a lot !