Hello, thank you for this code, i try to test imageDepthEstimation.py i did not find the Tensorflow models, can you please give me the link to download it . thank you

Hi,

I tried to run "python imageDepthEstimation.py" and encounter the following error:

NotImplementedError: Cannot convert a symbolic Tensor (PartitionedCall:0) to a numpy array.

when I traced the code below:

        def inference(self, input_tensor):
		output = self.model(input_tensor)
		return np.squeeze(output)

and found that: input_tensor: <tf.Tensor 'Const:0' shape=(1, 375, 450, 2) dtype=float32> output: <tf.Tensor 'PartitionedCall:0' shape=(?, ?, ?, ?) dtype=float32>

It seems the model is not working!? (I did trace the model path are correct and loading the model went through without any error message...)

Could you please help to shed some insight on what caussing this problem? and how to fix it?

environemnt: tensorflow: 1.15 python: 3.6.5 numpy: 1.19.5

Thanks.

Hi I'm trying to reproduce the metrics reported here https://github.com/zjjMaiMai/TinyHITNet

When running ModelType.middlebury on cones/im2.png cones/im6.png the disparity_map looks close in visualisation But it's values are maxed at ~55.0 and EPE metric is at ~98.0 when compared to cones/disp2.png

Could you please share the proper way to evaluate the model ?

errors:

/HITNET-Stereo-Depth-estimation/hitnet/utils_hitnet.py", line 28, in draw_disparity
    disparity_map = disparity_map.astype("uint8")
ValueError: setting an array element with a sequence.