C3D model conversion - Depth of input (112) is not a multiple of input depth of filter (3) for '{{node tf.nn.conv3d/Conv3D}}

tgcandido commented 1 year ago

Issue Type

Bug

OS

Ubuntu

OS architecture

x86_64

Programming Language

Python

Framework

TFLite

Download URL for ONNX / OpenVINO IR

https://pastebin.com/t0BqsAfG

Convert Script

openvino2tensorflow --model_path openvino/c3d-sim.xml --model_output_path ./openvino2tensorflow --output_saved_model --output_no_quant_float32_tflite

Description

Model C3D ( https://t.co/PeRc2UZI6x) was converted from Pytorch to ONNX, then to OpenVino, and finally to TFLite, but there is an error regarding dimensions. I followed this guide https://siliconlabs.github.io/mltk/mltk/tutorials/onnx_to_tflite.html with the C3D model, and the error was found in this specific section https://siliconlabs.github.io/mltk/mltk/tutorials/onnx_to_tflite.html#convert-from-openvino-to-tf-lite-float32.

Relevant Log Output

ERROR: Exception encountered when calling layer "tf.nn.conv3d" (type TFOpLambda).

Depth of input (112) is not a multiple of input depth of filter (3) for '{{node tf.nn.conv3d/Conv3D}} = Conv3D[T=DT_FLOAT, data_format="NDHWC", dilations=[1, 1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1, 1]](Placeholder, tf.nn.conv3d/Conv3D/filter)' with input shapes: [1,3,16,112,112], [64,3,3,3,3].

Source code for simple inference testing code

No response

PINTO0309 commented 1 year ago

First, read the README section and give it a try. It is almost a transposition of dimensions. https://github.com/PINTO0309/openvino2tensorflow#6-7-replace-weights-or-constant-values-in-const-op-and-add-transpose-or-reshape-or-cast-or-squeeze-or-unsqueeze-or-add-or-multiply-just-beforeafter-the-operation-specified-by-layer_id
Then, please describe in detail what you tried and how the problem occurred. Please describe all. I am not a support desk.

Miscellaneous issues are closed immediately.
If you have no intention of returning useful information to the community, you have no right to post an issue here.
Please withdraw.