Reported FLOPS metric incorrect for ConvNeXtSmall model

[isaacgerg]

The following code reports zero FLOPs which is incorrect.

import model_profiler
import tensorflow as tf
input_image = tf.keras.layers.Input(shape=(256, 256, 1))
rgb_layer = tf.keras.layers.concatenate([input_image, input_image, input_image]) * 255.0
ots = tf.keras.applications.ConvNeXtSmall(include_top=False, input_shape = (256, 256, 3), include_preprocessing=True, weights = init_weights)
fv = ots(rgb_layer)
f = tf.keras.layers.GlobalAveragePooling2D()(fv)
y_pred = tf.keras.layers.Dense(1, activation='sigmoid', name='output_class')(f)
model = tf.keras.models.Model(input_image, y_pred)
r = model_profiler.model_profiler(model, 1,  use_units = ['GPU IDs', 'MFLOPs', 'GB', 'Million', 'MB'])
print(r)

The above code prints:

| Model Profile                    | Value      | Unit    |
|----------------------------------|------------|---------|
| Selected GPUs                    | ['0', '1'] | GPU IDs |
| No. of FLOPs                     | 0.0        | MFLOPs  |
| GPU Memory Requirement           | 0.048      | GB      |
| Model Parameters                 | 49.4555    | Million |
| Memory Required by Model Weights | 188.6576   | MB      |

which is incorrect for the No. of FLOPs row.