README.md

DenseNet 169

This document has instructions for how to run DenseNet 169 for the following modes/precisions:

• FP32 inference

FP32 Inference Instructions

1. Download ImageNet dataset.

   This step is required only for running accuracy, for running the model for performance we do not need to provide dataset.

   Download and preprocess the ImageNet dataset using the instructions here. After running the conversion script you should have a directory with the ImageNet dataset in the TF records format.

2. Download the pretrained model:

   $ wget https://storage.googleapis.com/intel-optimized-tensorflow/models/v1_8/densenet169_fp32_pretrained_model.pb
   

3. Clone the intelai/models repo and then run the model scripts for either online or batch inference or accuracy. For --data-location in accuracy run, please use the ImageNet validation data path from step 1. Each model run has user configurable arguments separated from regular arguments by '--' at the end of the command. Unless configured, these arguments will run with default values. Below are the example codes for each use case:

   $ git clone https://github.com/IntelAI/models.git
   
   $ cd benchmarks
   

   For throughput (using --benchmark-only, --socket-id 0 and --batch-size 100):

   python launch_benchmark.py \
       --model-name densenet169 \
       --precision fp32 \
       --mode inference \
       --framework tensorflow \
       --benchmark-only \
       --batch-size 100 \
       --socket-id 0 \
       --in-graph /home/<user>/densenet169_fp32_pretrained_model.pb \
       --docker-image intel/intel-optimized-tensorflow:2.3.0 \
       -- input_height=224 input_width=224 warmup_steps=20 steps=100 \
       input_layer="input" output_layer="densenet169/predictions/Reshape_1"
   

   For latency (using --benchmark-only, --socket-id 0 and --batch-size 1)

   python launch_benchmark.py \
       --model-name densenet169 \
       --precision fp32 \
       --mode inference \
       --framework tensorflow \
       --benchmark-only \
       --batch-size 1 \
       --socket-id 0 \
       --in-graph /home/<user>/densenet169_fp32_pretrained_model.pb \
       --docker-image intel/intel-optimized-tensorflow:2.3.0 \
       -- input_height=224 input_width=224 warmup_steps=20 steps=100 \
       input_layer="input" output_layer="densenet169/predictions/Reshape_1"
   

   For accuracy (using your --data-location, --socket-id 0, --accuracy-only and --batch-size 100):

   python launch_benchmark.py \
       --model-name densenet169 \
       --precision fp32 \
       --mode inference \
       --framework tensorflow \
       --accuracy-only \
       --batch-size 100 \
       --socket-id 0 \
       --in-graph /home/<user>/densenet169_fp32_pretrained_model.pb \
       --docker-image intel/intel-optimized-tensorflow:2.3.0 \
       --data-location /home/<user>/imagenet_validation_dataset \
       -- input_height=224 input_width=224 \
       input_layer="input" output_layer="densenet169/predictions/Reshape_1"
   

   Note that the --verbose or --output-dir flag can be added to any of the above commands to get additional debug output or change the default output location.

4. The log file is saved to the models/benchmarks/common/tensorflow/logs directory, or the directory specified by the --output-dir arg. Below are examples of what the tail of your log file should look like for the different configs.

   Example log tail when running for batch inference:

   steps = 80, 159.83471377 images/sec
          Latency: 625.646317005 ms
   steps = 90, 159.852789241 images/sec
          Latency: 625.57557159 ms
   steps = 100, 159.853966416 images/sec
          Latency: 625.570964813 ms
   Ran inference with batch size 100
   Log location outside container: {--output-dir value}/benchmark_densenet169_inference_fp32_20190412_023940.log
   

   Example log tail when running for online inference:

   steps = 80, 34.9948442873 images/sec
          Latency: 28.5756379366 ms
   steps = 90, 34.9644341907 images/sec
          Latency: 28.6004914178 ms
   steps = 100, 34.9655988121 images/sec
          Latency: 28.5995388031 ms
   Ran inference with batch size 1
   Log location outside container: {--output-dir value}/benchmark_densenet169_inference_fp32_20190412_024505.log
   

   Example log tail when running for accuracy:

   Iteration time: 581.6446 ms
   0.757505030181
   Iteration time: 581.5755 ms
   0.757489959839
   Iteration time: 581.5709 ms
   0.75749498998
   Iteration time: 581.1705 ms
   0.75748
   Ran inference with batch size 100
   Log location outside container: {--output-dir value}/benchmark_densenet169_inference_fp32_20190412_021545.log