From 4349be1397c95b7fd19b8cebf586f1632f94d427 Mon Sep 17 00:00:00 2001 From: kevinlin311tw Date: Sat, 22 May 2021 18:12:40 -0700 Subject: [PATCH] update dataset downloads, and add logs --- docs/DOWNLOAD.md | 7 ++++--- docs/EXP.md | 3 +++ docs/INSTALL.md | 6 +++--- 3 files changed, 10 insertions(+), 6 deletions(-) diff --git a/docs/DOWNLOAD.md b/docs/DOWNLOAD.md index c9f0224..e97b04f 100644 --- a/docs/DOWNLOAD.md +++ b/docs/DOWNLOAD.md @@ -81,9 +81,10 @@ To download the annotation files, please use the following command. ```bash cd $REPO_DIR - path/to/azcopy copy 'https://datarelease.blob.core.windows.net/metro/datasets/datasets_release/DATA_NAME' --recursive + path/to/azcopy copy 'https://datarelease.blob.core.windows.net/metro/datasets/filename.tar' /path/to/your/folder/filename.tar + tar xvf filename.tar ``` - `DATA_NAME` could be `Tax-H36m-coco40k-Muco-UP-Mpii`, `human3.6m`, `coco_smpl`, `muco`, `up3d`, `mpii`, `3dpw`, `freihand`. Total file size is about 200 GB. + `filename.tar` could be `Tax-H36m-coco40k-Muco-UP-Mpii.tar`, `human3.6m.tar`, `coco_smpl.tar`, `muco.tar`, `up3d.tar`, `mpii.tar`, `3dpw.tar`, `freihand.tar`. Total file size is about 200 GB. The datasets and pseudo ground truth labels are provided by [Pose2Mesh](https://github.com/hongsukchoi/Pose2Mesh_RELEASE). We only reorganize the data format to better fit our training pipeline. We suggest to download the orignal image files from the offical dataset websites. @@ -154,4 +155,4 @@ |-- ... |-- ... - ``` \ No newline at end of file + ``` diff --git a/docs/EXP.md b/docs/EXP.md index dac3092..2769171 100644 --- a/docs/EXP.md +++ b/docs/EXP.md @@ -26,6 +26,8 @@ python -m torch.distributed.launch --nproc_per_node=8 \ --hidden_feat_dim 1024,256,64 ``` +Example training log can be found here [freihand_reproduce_log.txt]('https://datarelease.blob.core.windows.net/metro/models/freihand_reproduce_log.txt') + ### Testing After training, we use the final checkpoint (trained at 200 epoch) for testing. @@ -110,6 +112,7 @@ python -m torch.distributed.launch --nproc_per_node=8 \ --hidden_feat_dim 1024,256,128 ``` +Example training log can be found here [h36m_reproduce_log.txt]('https://datarelease.blob.core.windows.net/metro/models/h36m_reproduce_log.txt') ### Evaluation on Human3.6M diff --git a/docs/INSTALL.md b/docs/INSTALL.md index d8ae7bb..36d6dd0 100644 --- a/docs/INSTALL.md +++ b/docs/INSTALL.md @@ -33,8 +33,8 @@ pip install opendr matplotlib # Install METRO cd $INSTALL_DIR -git clone --recursive git@github.com:microsoft/METRO.git -cd METRO +git clone --recursive https://github.com/microsoft/MeshTransformer.git +cd MeshTransformer python setup.py build develop # Install requirements @@ -42,7 +42,7 @@ pip install -r requirements.txt # Install manopth cd $INSTALL_DIR -cd METRO +cd MeshTransformer pip install ./manopth/. unset INSTALL_DIR