inference_openvino.py low dice score

[malcolmping]

Hi experts,

after I follow the guidance to train model, I execute 2D inference_openvino.py. However, the result images in "inference_examples_openvino" folder are not correct (no any prediction region) and I'm wondering how can I fix this issue!

Thank you!

Malcolm

[ INFO ] Loading U-Net model to the plugin [ INFO ] Loading network files: ../openvino_models/FP32/saved_model.xml ../openvino_models/FP32/saved_model.bin [ INFO ] Loading model to the plugin [ INFO ] Image #22: Dice score = 1.0000 [ INFO ] Image #40: Dice score = 0.0011 [ INFO ] Image #56: Dice score = 0.0013 [ INFO ] Image #61: Dice score = 0.0029 [ INFO ] Image #400: Dice score = 0.0007 [ INFO ] Image #1100: Dice score = 1.0000 [ INFO ] Image #2229: Dice score = 0.0011 [ INFO ] Image #3136: Dice score = 0.0011 [ INFO ] Image #4385: Dice score = 0.0009 [ INFO ] Could not open font file /usr/share/fonts/truetype/noto/NotoColorEmoji.ttf: In FT2Font: Could not set the fontsize [ INFO ] generated new fontManager [ INFO ] Plotting the predictions and saving to png files. Please wait... Saved file: inference_examples_openvino/pred_group0 Saved file: inference_examples_openvino/pred_group1 Saved file: inference_examples_openvino/pred_group2

[tonyreina]

Did you train the model? I'm wondering if the results are ok if you run https://github.com/IntelAI/unet/blob/master/2D/plot_inference_examples.py with the original TF/Keras version. The OpenVINO version of the TF/Keras model should give you the same results as that.

[tonyreina]

Just wanted to check back to see if you were able to train the model? Did the training script report the final Keras/TF model Dice score?