From d1e09d204a1e787458171883f69cca11359929ed Mon Sep 17 00:00:00 2001 From: Sanjay_Uconn Date: Tue, 23 Feb 2021 16:31:03 -0800 Subject: [PATCH] fixed broken .md files (#149) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Co-authored-by: “sanjay” <“sanjay.vishwakarma@unity3d.com”> --- docs/source/Evaluation_Tutorial.md | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/docs/source/Evaluation_Tutorial.md b/docs/source/Evaluation_Tutorial.md index 5e8e32e0..a9d3ae9b 100644 --- a/docs/source/Evaluation_Tutorial.md +++ b/docs/source/Evaluation_Tutorial.md @@ -20,11 +20,11 @@ A larger dataset of 400k we used in our experiments can be made available [upon ### UnityGroceries-Real dataset -We've also made a new [dataset of 1.3k real images](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/UnityGroceriesReal.md) which contain groceries and corresponding bounding boxes. You can look at it if you wish, or simply [skip ahead](#part-2-train-a-model) if you're interested in training a model on this dataset. The test split of this dataset will be used in [part 3](#part-3-evaluate-a-model). +We've also made a new [dataset of 1.3k real images](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/UnityGroceriesReal.md#) which contain groceries and corresponding bounding boxes. You can look at it if you wish, or simply [skip ahead](#part-2-train-a-model) if you're interested in training a model on this dataset. The test split of this dataset will be used in [part 3](#part-3-evaluate-a-model). ### Create a new synthetic dataset using Unity Simulation (optional) -If you want to run the full end-to-end pipeline including synthetic dataset generation you can follow [this guide](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md) and then continue to run [this training pipeline](#train-on-synthetic-dataset-generated-on-unity-simulation). +If you want to run the full end-to-end pipeline including synthetic dataset generation you can follow [this guide](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md#) and then continue to run [this training pipeline](#train-on-synthetic-dataset-generated-on-unity-simulation). ## Part 2: Train a model on the UnityGroceries-SyntheticSample dataset @@ -111,7 +111,7 @@ You should specify the following parameters: Once the notebook server starts successfully, open the server and choose `SynthDet_Evaluation.ipynb` under `/datasetinsights/notebooks` directory. Follow the instructions in the notebook to visualize predictions and performance. -Alternatively, you can follow similar [instructions](RunningSynthDetCloud.md#step-6-run-dataset-statistics-using-the-datasetinsights-jupyter-notebook) to run notebooks on local host machine. Replace the first step with the following command: +Alternatively, you can use following command to run notebooks on local host machine. ```bash docker run \ @@ -183,7 +183,7 @@ Next you can jump to [part 3](#part-3-monitor-training-in-tensorboard) to monito ### Train on synthetic dataset generated on Unity Simulation -This section shows you how to train a model on your own dataset generated by running the [SynthDet] environment on [Unity Simulation](https://unity.com/products/unity-simulation). You can follow [these instructions](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md) to generate the dataset. +This section shows you how to train a model on your own dataset generated by running the [SynthDet] environment on [Unity Simulation](https://unity.com/products/unity-simulation). You can follow [these instructions](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md#) to generate the dataset. To train the model, simply import [**this pre-compiled pipeline**](https://raw.githubusercontent.com/Unity-Technologies/datasetinsights/0.2.x/kubeflow/compiled/train_on_synthetic_dataset_unity_simulation.yaml) into your kubeflow cluster. The figure below shows how to do this using the [web UI](https://www.kubeflow.org/docs/pipelines/pipelines-quickstart/#deploy-kubeflow-and-open-the-pipelines-ui). You can optionally use the [KFP CLI Tool](https://www.kubeflow.org/docs/pipelines/sdk/sdk-overview/#kfp-cli-tool). @@ -202,7 +202,7 @@ You have to specify run parameters required by this pipeline: - `config`: Estimator config YAML file. You can use the default value which points to a YAML file packaged with our docker images or you can load from remote locations GCS or any HTTP(s) using file prefix `gs://, http(s)://`. - `tb_log_dir`: Path to store tensorboard logs used to visualize the training progress. - `checkpoint_dir`: Path to store output Estimator checkpoints. You can use one of the checkpoints for estimator evaluation. -- `volume_size`: Size of the Kubernetes Persistent Volume Claims (PVC) that will be used to store the dataset. You should change this value according to the dataset that was generated. If you use default settings from [these instructions](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md), you should expect `1.2TiB` storage required for 400k images. +- `volume_size`: Size of the Kubernetes Persistent Volume Claims (PVC) that will be used to store the dataset. You should change this value according to the dataset that was generated. If you use default settings from [these instructions](https://github.com/Unity-Technologies/SynthDet/blob/master/docs/RunningSynthDetCloud.md#), you should expect `1.2TiB` storage required for 400k images. Set `tb_log_dir` and `checkpoint_dir` to a location that is convenient for you and your Kubernetes cluster has permissions to write to. This is typically a GCS path under the same GCP project. You want to keep a note on these directories that will be used for tensorboard visualization and model evaluation. Note that an invalid location will cause the job to fail, whereas a path to the local filesystem may run but will be hard to monitor as you won't have easy access to the files.