rogerhcheng / liteflownet2-tf2 Goto Github PK

Do you guys see any issues converting the model to tflite for running on a coral chip?
I can work on the conversion if you think it's possible

I was looking to use this model for temporal extractor between frames in image translation. frankly im pretty new to TensorFlow. 
im trying to pass two tensor variables for LiteflowNet() class but this raise is only supported inside of tf.function or when eager execution is enabled.

Using the same command as in the README:
python eval.py --img1=./images/first.png --img2=./images/second.png --use_Sintel=True --display_flow=True --img_out=out.png

I get a flow result with all nans. Why?

Same for --use_Sintel=False. Basically the same result as this LiteFlowNet issue: keeper121/liteflownet-tf2#7

I am running TensorFlow 2.2.0, addons 0.10.0, Python 3.8.5, Pillow 8.0.1.

I get "InvalidArgumentError: Computed input depth 2 doesn't match filter input depth 1 [Op:Conv2DBackpropInput]"
in def group_upconv(), when running the code

Traceback (most recent call last):
File "eval.py", line 72, in
flow_color = flow_to_color(flow, convert_to_bgr=False)
File "D:\Dataset\LiteFlowNet2-TF2-master\draw_flow.py", line 124, in flow_to_color
return flow_compute_color(u, v, convert_to_bgr)
File "D:\Dataset\LiteFlowNet2-TF2-master\draw_flow.py", line 83, in flow_compute_color
col0 = tmp[k0] / 255.0
IndexError: index -2147483648 is out of bounds for axis 0 with size 55