2 errores in Custom Prediction

[SMY92]

Recently, I'm trying to train my own custom prediction model using ImageAI. The training process seems good. However, when I tried to run prediction with them, none of them could work. For the model SqueezeNet, there is only one problem which is "ValueError: Shapes (1, 1, 512, 3) and (2, 512, 1, 1) are incompatible". For the other three models, there is one more problem which is "ValueError: You have specified an incorrect path to the DenseNet model file.". The code I used is quit simple. En the case of SqueezeNet: Here is the code for training:

import os

os.environ["CUDA_VISIBLE_DEVICES"]="-1"

from imageai.Prediction.Custom import ModelTraining model_trainer = ModelTraining() model_trainer.setModelTypeAsSqueezeNet() model_trainer.setDataDirectory("Samples") model_trainer.trainModel(num_objects=2, num_experiments=30, enhance_data=True, batch_size=64, show_network_summary=True)

Here is the code for prediction:

from imageai.Prediction.Custom import CustomImagePrediction import os

execution_path = os.getcwd()

prediction = CustomImagePrediction() prediction.setModelTypeAsResNet() prediction.setModelPath(os.path.join(execution_path, "model_ex-030_acc-0.500000.h5")) prediction.setJsonPath(os.path.join(execution_path, "model_class.json")) prediction.loadModel()

predictions, probabilities = prediction.predictImage(os.path.join(execution_path, "image1.jpg"))

for eachPrediction, eachProbability in zip(predictions, probabilities): print(eachPrediction , " : " , eachProbability)

[OlafenwaMoses]

If you you trained your network with SqueezeNet, you must set the model type as SqueezeNet too in your CustomImagePrediction.

You training and prediction network must be the same.

[SMY92]

Sorry. I posted the wrong code. Do you means to write 'prediction.setModelTypeAsSqueezesNet()' instead of 'prediction.setModelTypeAsResNet()'? If you do, unfortunately, I've already done that.

[pytesnim]

My confusion is I don't know when 'model_ex-030_acc-0.500000.h5' is created? Do we download it from somewhere? Or is it the output of custom training? Thanks very much!

[OlafenwaMoses]

@pytesnim In the folder where your dataset is contained, check for the folder with the name models . You will find all the generated model in there.

[tryinggo]

Why Report an error "raise ValueError("You have specified an incorrect path to the ResNet model file.") ValueError: You have specified an incorrect path to the ResNet model file." ,For custom image prediction

[OlafenwaMoses]

Hello @SMY92 @pytesnim @tryinggo , thank you all for your patience. This issue has been resolved in the latest version (v2.0.3). See the link below for details on this.

https://github.com/OlafenwaMoses/ImageAI/blob/master/imageai/Prediction/CUSTOMPREDICTION.md#custompredictionmultiple