Loading Model from Keras

[wgcv]

TensorFlow.js version

1.1.2

Browser version

Chrome 75.0.3770.100 (Official Build) (64-bit)

Tensorflow

Tensorflow 2.0 Beta

Describe the problem or feature request

I export my trained model (Work great with Tensorflow 2.0 and Keras)

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_4 (Conv2D)            (None, 28, 28, 32)        160       
_________________________________________________________________
leaky_re_lu (LeakyReLU)      (None, 28, 28, 32)        0         
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 13, 13, 64)        8256      
_________________________________________________________________
leaky_re_lu_1 (LeakyReLU)    (None, 13, 13, 64)        0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 13, 13, 64)        0         
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 6, 6, 64)          0         
_________________________________________________________________
dropout_5 (Dropout)          (None, 6, 6, 64)          0         
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 5, 5, 196)         50372     
_________________________________________________________________
leaky_re_lu_2 (LeakyReLU)    (None, 5, 5, 196)         0         
_________________________________________________________________
dropout_6 (Dropout)          (None, 5, 5, 196)         0         
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 2, 2, 196)         0         
_________________________________________________________________
dropout_7 (Dropout)          (None, 2, 2, 196)         0         
_________________________________________________________________
global_max_pooling2d (Global (None, 196)               0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 196)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 1024)              201728    
_________________________________________________________________
leaky_re_lu_3 (LeakyReLU)    (None, 1024)              0         
_________________________________________________________________
dropout_8 (Dropout)          (None, 1024)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 62)                63550     
=================================================================
Total params: 324,066
Trainable params: 324,066
Non-trainable params: 0
________________________

with

import tensorflowjs as tfjs
tfjs.converters.save_keras_model(model, "./js")

but when i load with

  <script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@1.1.2/dist/tf.min.js"></script>
<script type="text/javascript">
        const model = await tf.loadLayersModel('model/model.json');
</script>

I get this error

Uncaught (in promise) Error: Input 0 is incompatible with layer flatten_1: expected min_ndim=3, found ndim=2. Error: Input 0 is incompatible with layer flatten_1: expected min_ndim=3, found ndim=2.
    at new e (errors.ts:48)
    at e.assertInputCompatibility (topology.ts:789)
    at topology.ts:984
    at Yp (common.ts:43)
    at e.apply (topology.ts:977)
    at e.add (models.ts:499)
    at e.fromConfig (models.ts:1020)
    at Cp (generic_utils.ts:277)
    at ud (serialization.ts:31)
    at models.ts:299
e @ errors.ts:48
e.assertInputCompatibility @ topology.ts:789
(anonymous) @ topology.ts:984
Yp @ common.ts:43
e.apply @ topology.ts:977
e.add @ models.ts:499
e.fromConfig @ models.ts:1020
Cp @ generic_utils.ts:277
ud @ serialization.ts:31
(anonymous) @ models.ts:299
(anonymous) @ common.ts:14
(anonymous) @ common.ts:14
o @ common.ts:14
errors.ts:48 Uncaught (in promise) Error: Input 0 is incompatible with layer flatten_1: expected min_ndim=3, found ndim=2.
    at new e (errors.ts:48)
    at e.assertInputCompatibility (topology.ts:789)
    at topology.ts:984
    at Yp (common.ts:43)
    at e.apply (topology.ts:977)
    at e.add (models.ts:499)
    at e.fromConfig (models.ts:1020)
    at Cp (generic_utils.ts:277)
    at ud (serialization.ts:31)
    at models.ts:299
e @ errors.ts:48
e.assertInputCompatibility @ topology.ts:789
(anonymous) @ topology.ts:984
Yp @ common.ts:43
e.apply @ topology.ts:977
e.add @ models.ts:499
e.fromConfig @ models.ts:1020
Cp @ generic_utils.ts:277
ud @ serialization.ts:31
(anonymous) @ models.ts:299
(anonymous) @ common.ts:14
(anonymous) @ common.ts:14
o @ common.ts:14

Code to reproduce the bug / link to feature request

[wgcv]

I solved the problem removing this lines:

                    {
                        "class_name": "Flatten",
                        "config": {
                            "name": "flatten_1",
                            "trainable": true,
                            "dtype": "float32",
                            "data_format": "channels_last"
                        }

But i don't think that this is the best solution, this is a error of the export tool or in TF.Js?

[tafsiri]

cc @caisq @pyu10055 could you take a look.

[rthadur]

Automatically closing due to lack of recent activity. Please update the issue when new information becomes available, and we will reopen the issue. Thanks!