Skip to content

GitLab

Unverified Commit 5fd32ef6 authored by Suharsh Sivakumar's avatar Suharsh Sivakumar Committed by GitHub
Browse files

Merge pull request #4790 from suharshs/master 

Update mobilenet_v1 docs to include revised tflite quantized models.
parents 308179f3 a0db2413
Hide whitespace changes
Inline Side-by-side
Showing with 36 additions and 32 deletions
research/slim/nets/mobilenet_v1.md
View file @ 5fd32ef6
...@@ -17,38 +17,42 @@ image classification dataset. Accuracies were computed by evaluating using a sin ...@@ -17,38 +17,42 @@ image classification dataset. Accuracies were computed by evaluating using a sin
Model | Million MACs | Million Parameters | Top-1 Accuracy| Top-5 Accuracy | Model | Million MACs | Million Parameters | Top-1 Accuracy| Top-5 Accuracy |
:----:|:------------:|:----------:|:-------:|:-------:| :----:|:------------:|:----------:|:-------:|:-------:|
[MobileNet_v1_1.0_224](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_224.tgz)|569|4.24|70.9|89.9| [MobileNet_v1_1.0_224](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224.tgz)|569|4.24|70.9|89.9|
[MobileNet_v1_1.0_192](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_192.tgz)|418|4.24|70.0|89.2| [MobileNet_v1_1.0_192](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_192.tgz)|418|4.24|70.0|89.2|
[MobileNet_v1_1.0_160](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_160.tgz)|291|4.24|68.0|87.7| [MobileNet_v1_1.0_160](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_160.tgz)|291|4.24|68.0|87.7|
[MobileNet_v1_1.0_128](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_128.tgz)|186|4.24|65.2|85.8| [MobileNet_v1_1.0_128](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_128.tgz)|186|4.24|65.2|85.8|
[MobileNet_v1_0.75_224](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_224.tgz)|317|2.59|68.4|88.2| [MobileNet_v1_0.75_224](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_224.tgz)|317|2.59|68.4|88.2|
[MobileNet_v1_0.75_192](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_192.tgz)|233|2.59|67.2|87.3| [MobileNet_v1_0.75_192](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_192.tgz)|233|2.59|67.2|87.3|
[MobileNet_v1_0.75_160](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_160.tgz)|162|2.59|65.3|86.0| [MobileNet_v1_0.75_160](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_160.tgz)|162|2.59|65.3|86.0|
[MobileNet_v1_0.75_128](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_128.tgz)|104|2.59|62.1|83.9| [MobileNet_v1_0.75_128](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_128.tgz)|104|2.59|62.1|83.9|
[MobileNet_v1_0.50_224](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_224.tgz)|150|1.34|63.3|84.9| [MobileNet_v1_0.50_224](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_224.tgz)|150|1.34|63.3|84.9|
[MobileNet_v1_0.50_192](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_192.tgz)|110|1.34|61.7|83.6| [MobileNet_v1_0.50_192](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_192.tgz)|110|1.34|61.7|83.6|
[MobileNet_v1_0.50_160](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_160.tgz)|77|1.34|59.1|81.9| [MobileNet_v1_0.50_160](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_160.tgz)|77|1.34|59.1|81.9|
[MobileNet_v1_0.50_128](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_128.tgz)|49|1.34|56.3|79.4| [MobileNet_v1_0.50_128](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_128.tgz)|49|1.34|56.3|79.4|
[MobileNet_v1_0.25_224](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_224.tgz)|41|0.47|49.8|74.2| [MobileNet_v1_0.25_224](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_224.tgz)|41|0.47|49.8|74.2|
[MobileNet_v1_0.25_192](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_192.tgz)|34|0.47|47.7|72.3| [MobileNet_v1_0.25_192](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_192.tgz)|34|0.47|47.7|72.3|
[MobileNet_v1_0.25_160](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_160.tgz)|21|0.47|45.5|70.3| [MobileNet_v1_0.25_160](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_160.tgz)|21|0.47|45.5|70.3|
[MobileNet_v1_0.25_128](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_128.tgz)|14|0.47|41.5|66.3| [MobileNet_v1_0.25_128](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_128.tgz)|14|0.47|41.5|66.3|
[MobileNet_v1_1.0_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_224_quant.tgz)|569|4.24|69.7|89.5| [MobileNet_v1_1.0_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz)|569|4.24|70.1|88.9|
[MobileNet_v1_1.0_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_192_quant.tgz)|418|4.24|69.0|88.9| [MobileNet_v1_1.0_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_192_quant.tgz)|418|4.24|69.2|88.3|
[MobileNet_v1_1.0_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_160_quant.tgz)|291|4.24|67.3|87.7| [MobileNet_v1_1.0_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_160_quant.tgz)|291|4.24|67.2|86.7|
[MobileNet_v1_1.0_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_128_quant.tgz)|186|4.24|64.0|85.5| [MobileNet_v1_1.0_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_128_quant.tgz)|186|4.24|63.4|84.2|
[MobileNet_v1_0.75_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_224_quant.tgz)|317|2.59|67.9|88.1| [MobileNet_v1_0.75_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_224_quant.tgz)|317|2.59|66.8|87.0|
[MobileNet_v1_0.75_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_192_quant.tgz)|233|2.59|66.2|87.1| [MobileNet_v1_0.75_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_192_quant.tgz)|233|2.59|66.1|86.4|
[MobileNet_v1_0.75_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_160_quant.tgz)|162|2.59|63.9|85.5| [MobileNet_v1_0.75_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_160_quant.tgz)|162|2.59|62.3|83.8|
[MobileNet_v1_0.75_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.75_128_quant.tgz)|104|2.59|59.8|82.8| [MobileNet_v1_0.75_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_128_quant.tgz)|104|2.59|55.8|78.8|
[MobileNet_v1_0.50_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_224_quant.tgz)|150|1.34|62.2|84.5| [MobileNet_v1_0.50_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_224_quant.tgz)|150|1.34|60.7|83.2|
[MobileNet_v1_0.50_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_192_quant.tgz)|110|1.34|60.4|83.2| [MobileNet_v1_0.50_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_192_quant.tgz)|110|1.34|60.0|82.2|
[MobileNet_v1_0.50_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_160_quant.tgz)|77|1.34|57.7|81.3| [MobileNet_v1_0.50_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_160_quant.tgz)|77|1.34|57.7|80.4|
[MobileNet_v1_0.50_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.5_128_quant.tgz)|49|1.34|54.9|78.9| [MobileNet_v1_0.50_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.5_128_quant.tgz)|49|1.34|54.5|77.7|
[MobileNet_v1_0.25_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_224_quant.tgz)|41|0.47|48.2|73.8| [MobileNet_v1_0.25_224_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_224_quant.tgz)|41|0.47|48.0|72.8|
[MobileNet_v1_0.25_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_192_quant.tgz)|34|0.47|45.8|71.9| [MobileNet_v1_0.25_192_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_192_quant.tgz)|34|0.47|46.0|71.2|
[MobileNet_v1_0.25_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_160_quant.tgz)|21|0.47|43.5|69.1| [MobileNet_v1_0.25_160_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_160_quant.tgz)|21|0.47|43.4|68.5|
[MobileNet_v1_0.25_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_0.25_128_quant.tgz)|14|0.47|39.9|65.8| [MobileNet_v1_0.25_128_quant](http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.25_128_quant.tgz)|14|0.47|39.5|64.4|
Revisions to models:
* July 12, 2018: Update to TFLite models that fixes an accuracy issue resolved by making conversion support weights with narrow_range. We now report validation on the actual TensorFlow Lite model rather than the emulated quantization number of TensorFlow.
* August 2, 2018: Update to TFLite models that fixes an accuracy issue resolved by making sure the numerics of quantization match TF quantized training accurately.
The linked model tar files contain the following: The linked model tar files contain the following:
* Trained model checkpoints * Trained model checkpoints
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment