Create README.md for semantic segmentation

Author: rod409

automotive/semantic-segmentation/README.md

@@ -0,0 +1,32 @@
+# ABTF Semantic Segmentation Benchmark
+
+This is the reference implementation for the ABTF semantic segmentation benchmark. The reference uses ONNX as a backend. A pytorch implementation is provided as well.
+
+| model | accuracy | dataset | model source | precision |
+| ---- | ---- | ---- | ---- | ---- |
+| DeepLabv3+ | TBD | Cognata | https://github.com/rod409/pp/tree/main/deeplabv3plus | fp32 |
+
+## Downloading the dataset and model checkpoints
+Contact [MLCommons](https://mlcommons.org/datasets/cognata) to access the cognata dataset. Access requires MLCommons membership and signing the EULA. The dataset download also contains the DeepLabv3+ onnx and pytorch model checkpoints.
+After downloading the datasets extract the compressed files.
+
+You will need to perform some dataset preprocessing for segmentation. Follow the instructions on the [source repo](https://github.com/rod409/pp/tree/main/deeplabv3plus) to process the dataset.
+
+> [!Note]
+> Providing preprocessed data is in progress.
+
+## Build and run the Docker container
+```
+git clone -b v0.5abtf [email protected]:rod409/inference.git
+cd inference/automotive/semantic-segmentation
+docker build -t deeplab_inference .
+docker run -it -v ./inference:/inference -v <path to cognata>:/cognata deeplab_inference
+```
+## Run the model in performance mode
+```
+python main.py --backend onnx --checkpoint /cognata/deeplabv3+.onnx --dataset-path /cognata/ --dataset cognata
+```
+## Run the model in accuracy mode and run the accuracy checker
+```
+python main.py --backend onnx --checkpoint /cognata/deeplabv3+.onnx --dataset-path /cognata/ --dataset cognata --accuracy
+python accuracy_cognata.py --mlperf-accuracy-file ./output/mlperf_log_accuracy.json --dataset-path /Documents/research/datasets/cognata/