Shape mismatch in a DenseNet121 converted from pytorch through keras. #824
Description
TensorFlow.js version
0.13.2
Browser version
Firefox 62.0.3 64-bit
Chrome Version 68.0.3440.84 (Official Build) (64-bit)
Ubuntu 18.04
Describe the problem or feature request
DenseNet converted from pytorch to keras and then from keras to tfjs throws an error on load in tensorflow-js:
Error: Shape mismatch: [null,1000] vs. [1024,1000]
Code to reproduce the bug / link to feature request
Loading the model on the js side:
import * as tf from '@tensorflow/tfjs';
mobilenet = await tf.loadModel('https://s3-eu-west-1.amazonaws.com/kidzinski/densenet121/model.json');With package.json:
{
"name": "tfjs-examples-mobilenet",
"version": "0.1.0",
"description": "",
"main": "index.js",
"license": "Apache-2.0",
"private": true,
"engines": {
"node": ">=8.9.0"
},
"dependencies": {
"@tensorflow/tfjs": "^0.13.2",
"express": "^4.16.4"
},
"scripts": {
"watch": "cross-env NODE_ENV=development parcel index.html --no-hmr --open",
"build": "cross-env NODE_ENV=production parcel build index.html --no-minify --public-url ./",
"link-local": "yalc link",
"assets": "cp -R public/ dist/public/",
"postinstall": "yarn upgrade --pattern @tensorflow"
},
"devDependencies": {
"babel-plugin-transform-runtime": "~6.23.0",
"babel-polyfill": "~6.26.0",
"babel-preset-env": "~1.6.1",
"clang-format": "~1.2.2",
"cross-env": "^5.1.6",
"parcel-bundler": "~1.8.1",
"yalc": "~1.0.0-pre.22"
}
}Conversion from pytorch to keras (how I created the model under the link above)
import torchvision
import numpy as np
import tensorflowjs as tfjs
import torch
from torch.autograd import Variable
from pytorch2keras.converter import pytorch_to_keras
pretrained_model = torchvision.models.densenet121(pretrained = True)
pretrained_model = pretrained_model.float()
img_size = 224
input_np = np.random.uniform(0, 1, (1, 3, img_size, img_size))
input_dummy = Variable(torch.FloatTensor(input_np))
k_model = pytorch_to_keras(pretrained_model, input_dummy, [(3, img_size, img_size,)], verbose=True, names="short")
tfjs.converters.save_keras_model(k_model, "DenseNet121")With conda environment
# packages in environment at /home/kidzik/anaconda3/envs/tensorflow:
#
# Name Version Build Channel
blas 1.0 mkl
ca-certificates 2018.03.07 0
certifi 2018.10.15 py36_0
cffi 1.11.5 py36he75722e_1
cuda92 1.0 0 pytorch
cycler 0.10.0 <pip>
dill 0.2.8.2 <pip>
freetype 2.9.1 h8a8886c_1
h5py 2.8.0 <pip>
intel-openmp 2019.0 118
jpeg 9b h024ee3a_2
jupyter 1.0.0 <pip>
Keras 2.2.2 <pip>
Keras-Applications 1.0.4 <pip>
Keras-Preprocessing 1.0.2 <pip>
kiwisolver 1.0.1 <pip>
libedit 3.1.20170329 h6b74fdf_2
libffi 3.2.1 hd88cf55_4
libgcc-ng 8.2.0 hdf63c60_1
libgfortran-ng 7.3.0 hdf63c60_0
libpng 1.6.35 hbc83047_0
libstdcxx-ng 8.2.0 hdf63c60_1
libtiff 4.0.9 he85c1e1_2
matplotlib 3.0.0 <pip>
mkl 2019.0 118
mkl_fft 1.0.6 py36h7dd41cf_0
mkl_random 1.0.1 py36h4414c95_1
ncurses 6.1 hf484d3e_0
ninja 1.8.2 py36h6bb024c_1
numpy 1.15.3 py36h1d66e8a_0
numpy 1.14.1 <pip>
numpy-base 1.15.3 py36h81de0dd_0
olefile 0.46 py36_0
onnx 1.3.0 <pip>
onnx-tf 1.2.0 <pip>
opencv-python 3.4.3.18 <pip>
openssl 1.1.1 h7b6447c_0
pillow 5.3.0 py36h34e0f95_0
pip 10.0.1 py36_0
pycparser 2.19 py36_0
pyparsing 2.2.2 <pip>
python 3.6.7 h0371630_0
pytorch 0.4.1 py36_cuda9.2.148_cudnn7.1.4_1 [cuda92] pytorch
pytorch2keras 0.1.5 <pip>
PyYAML 3.13 <pip>
readline 7.0 h7b6447c_5
scikit-learn 0.20.0 <pip>
scipy 1.1.0 <pip>
setuptools 39.1.0 <pip>
setuptools 40.4.3 py36_0
six 1.11.0 py36_1
sqlite 3.25.2 h7b6447c_0
tensorboard 1.9.0 <pip>
tensorflow 1.9.0 <pip>
tensorflow-hub 0.1.1 <pip>
tensorflowjs 0.6.4 <pip>
tk 8.6.8 hbc83047_0
torch 0.4.0 <pip>
torchvision 0.2.1 py36_1 pytorch
typing 3.6.6 <pip>
typing-extensions 3.6.6 <pip>
wheel 0.32.2 py36_0
xz 5.2.4 h14c3975_4
zlib 1.2.11 ha838bed_2