Tflite validation

docker/TensorFlowLite/Dockerfile

@@ -0,0 +1,29 @@
+FROM ubuntu_for_dli
+
+WORKDIR /root/
+ARG TFLITE_VERSION=2.10.0
+ARG TF_VERSION=2.13.0
+RUN pip3 install --upgrade pip && \
+    pip3 install opencv-python requests PyYAML docker
+RUN export LD_LIBRARY_PATH=/root/miniconda3/lib:${LD_LIBRARY_PATH}
+
+WORKDIR /tmp/
+COPY models.lst models.lst
+RUN python3 ./open_model_zoo/tools/model_tools/downloader.py --name mobilenet-v1-1.0-224-tf
+
+RUN pip3 install tensorflow==${TF_VERSION}
+RUN pip3 install tflite==${TFLITE_VERSION}
+RUN pip3 install -r ./dl-benchmark/src/model_converters/requirements.txt
+RUN python3 ./open_model_zoo/tools/model_tools/converter.py --name mobilenet-v1-1.0-224-tf
+RUN rm models.lst
+
+WORKDIR /tmp/open_model_zoo/tools/accuracy_checker
+COPY config_tf2.yml config_tf2.yml
+
+
+RUN python3 setup.py install_core
+RUN accuracy_check -c config_tf2.yml -m ../../../public -s sample
+
+RUN rm config_tf2.yml
+
+WORKDIR /tmp/

docker/TensorFlowLite/config_tf2.yml

@@ -0,0 +1,31 @@
+models:
+  - name: mobilenet-v1-1.0-224-tf
+    launchers:
+      - framework: tf_lite
+        device: CPU
+        model:   mobilenet-v1-1.0-224-tf/mobilenet_v1_1.0_224.tflite
+        saved_model_dir: mobilenet-v1-1.0-224-tf/
+        adapter: classification
+        batch: 1
+
+    datasets:
+      - name: sample_dataset
+        data_source: sample_dataset/test
+        annotation_conversion:
+          converter: cifar
+          data_batch_file: cifar-10-batches-py/test_batch
+          convert_images: True
+          converted_images_dir: sample_dataset/test
+          num_classes: 10
+
+        preprocessing:
+          - type: resize
+            size: 224
+          - type: bgr_to_rgb
+          - type: normalization
+            mean: (127.5, 127.5, 127.5)
+            std: (127.5, 127.5, 127.5)
+
+        metrics:
+          - type: accuracy
+            top_k: 1

results/tflite_models_checklist.md

@@ -7,9 +7,9 @@
 Model | Availability in OMZ (2023.03.10) | Availability in the validation table |
 -|-|-|
 densenet-121|+-|+|
-googlenet-v1-tf|+-|-|
-googlenet-v2-tf|+-|-|
-googlenet-v3|+-|-|
+googlenet-v1-tf|+-|+|
+googlenet-v2-tf|+-|+|
+googlenet-v3|+-|+|
 googlenet-v4-tf|+-|+|
 mobilenet-v1-1.0-224-tf|+-|+|
 mobilenet-v2-1.0-224|+-|+|
@@ -47,21 +47,21 @@ lite-model_mobilenet_v1_100_224_uint8_1.tflite|+ ([link][mobilenet_v1_100_224_ui
 lite-model_mobilenet_v2_100_224_uint8_1.tflite|+ ([link][mobilenet_v2_100_224_uint8_1])|+|
 lite-model_mobilenet_v3_small_100_224_uint8_1.tflite|+ ([link][mobilenet_v3_small_100_224_uint8_1])|+|
 lite-model_mobilenet_v3_large_100_224_uint8_1.tflite|+ ([link][mobilenet_v3_large_100_224_uint8_1])|+|
-efficientnet_lite0_fp32_2.tflite|+ ([link][efficientnet_lite0_fp32_2])|-|
-efficientnet_lite1_fp32_2.tflite|+ ([link][efficientnet_lite1_fp32_2])|-|
-efficientnet_lite2_fp32_2.tflite|+ ([link][efficientnet_lite2_fp32_2])|-|
-efficientnet_lite3_fp32_2.tflite|+ ([link][efficientnet_lite3_fp32_2])|-|
-efficientnet_lite4_fp32_2.tflite|+ ([link][efficientnet_lite4_fp32_2])|-|
-lite-efficientnet_lite0_uint8_2.tflite|+ ([link][efficientnet_lite0_uint8_2])|-|
-lite-efficientnet_lite1_uint8_2.tflite|+ ([link][efficientnet_lite1_uint8_2])|-|
-lite-efficientnet_lite2_uint8_2.tflite|+ ([link][efficientnet_lite2_uint8_2])|-|
-lite-efficientnet_lite3_uint8_2.tflite|+ ([link][efficientnet_lite3_uint8_2])|-|
-lite-efficientnet_lite4_uint8_2.tflite|+ ([link][efficientnet_lite4_uint8_2])|-|
-efficientnet_lite0_int8_2.tflite|+ ([link][efficientnet_lite0_int8_2])|-|
-efficientnet_lite1_int8_2.tflite|+ ([link][efficientnet_lite1_int8_2])|-|
-efficientnet_lite2_int8_2.tflite|+ ([link][efficientnet_lite2_int8_2])|-|
-efficientnet_lite3_int8_2.tflite|+ ([link][efficientnet_lite3_int8_2])|-|
-efficientnet_lite4_int8_2.tflite|+ ([link][efficientnet_lite4_int8_2])|-|
+efficientnet_lite0_fp32_2.tflite|+ ([link][efficientnet_lite0_fp32_2])|+|
+efficientnet_lite1_fp32_2.tflite|+ ([link][efficientnet_lite1_fp32_2])|+|
+efficientnet_lite2_fp32_2.tflite|+ ([link][efficientnet_lite2_fp32_2])|+|
+efficientnet_lite3_fp32_2.tflite|+ ([link][efficientnet_lite3_fp32_2])|+|
+efficientnet_lite4_fp32_2.tflite|+ ([link][efficientnet_lite4_fp32_2])|+|
+lite-efficientnet_lite0_uint8_2.tflite|+ ([link][efficientnet_lite0_uint8_2])|+|
+lite-efficientnet_lite1_uint8_2.tflite|+ ([link][efficientnet_lite1_uint8_2])|+|
+lite-efficientnet_lite2_uint8_2.tflite|+ ([link][efficientnet_lite2_uint8_2])|+|
+lite-efficientnet_lite3_uint8_2.tflite|+ ([link][efficientnet_lite3_uint8_2])|+|
+lite-efficientnet_lite4_uint8_2.tflite|+ ([link][efficientnet_lite4_uint8_2])|+|
+efficientnet_lite0_int8_2.tflite|+ ([link][efficientnet_lite0_int8_2])|+|
+efficientnet_lite1_int8_2.tflite|+ ([link][efficientnet_lite1_int8_2])|+|
+efficientnet_lite2_int8_2.tflite|+ ([link][efficientnet_lite2_int8_2])|+|
+efficientnet_lite3_int8_2.tflite|+ ([link][efficientnet_lite3_int8_2])|+|
+efficientnet_lite4_int8_2.tflite|+ ([link][efficientnet_lite4_int8_2])|+|
 
 **Note:** inference implementation for EfficientNet-models
 supported for batch size that equals 1.