Merge branch 'MetaTensor_Spacing' into MetaTensor

Author: wyli

.github/workflows/pep8.yml

@@ -4,7 +4,7 @@ on:
   # quick tests for every pull request
   push:
     branches:
-      - master
+      - main
   pull_request:
 
 jobs:

3d_segmentation/brats_segmentation_3d.ipynb

@@ -267,13 +267,13 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=\"image\"),\n",
     "        Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
-    "        ConvertToMultiChannelBasedOnBratsClassesd(keys=\"label\"),\n",
     "        Spacingd(\n",
     "            keys=[\"image\", \"label\"],\n",
     "            pixdim=(1.0, 1.0, 1.0),\n",
     "            mode=(\"bilinear\", \"nearest\"),\n",
     "        ),\n",
+    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
+    "        ConvertToMultiChannelBasedOnBratsClassesd(keys=\"label\"),\n",
     "        RandSpatialCropd(keys=[\"image\", \"label\"], roi_size=[224, 224, 144], random_size=False),\n",
     "        RandFlipd(keys=[\"image\", \"label\"], prob=0.5, spatial_axis=0),\n",
     "        RandFlipd(keys=[\"image\", \"label\"], prob=0.5, spatial_axis=1),\n",
@@ -289,13 +289,13 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=\"image\"),\n",
     "        Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
-    "        ConvertToMultiChannelBasedOnBratsClassesd(keys=\"label\"),\n",
     "        Spacingd(\n",
     "            keys=[\"image\", \"label\"],\n",
     "            pixdim=(1.0, 1.0, 1.0),\n",
     "            mode=(\"bilinear\", \"nearest\"),\n",
     "        ),\n",
+    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
+    "        ConvertToMultiChannelBasedOnBratsClassesd(keys=\"label\"),\n",
     "        NormalizeIntensityd(keys=\"image\", nonzero=True, channel_wise=True),\n",
     "        EnsureTyped(keys=[\"image\", \"label\"]),\n",
     "    ]\n",
@@ -788,9 +788,9 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=[\"image\"]),\n",
     "        Orientationd(keys=[\"image\"], axcodes=\"RAS\"),\n",
+    "        Spacingd(keys=[\"image\"], pixdim=(1.0, 1.0, 1.0), mode=\"bilinear\"),\n",
     "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        ConvertToMultiChannelBasedOnBratsClassesd(keys=\"label\"),\n",
-    "        Spacingd(keys=[\"image\"], pixdim=(1.0, 1.0, 1.0), mode=\"bilinear\"),\n",
     "        NormalizeIntensityd(keys=\"image\", nonzero=True, channel_wise=True),\n",
     "        EnsureTyped(keys=[\"image\", \"label\"]),\n",
     "    ]\n",

3d_segmentation/challenge_baseline/README.md

@@ -109,9 +109,9 @@ This baseline method achieves 0.6904 ± 0.1801 Dice score on the challenge valid
 
 - For MONAI technical documentation, please visit [docs.monai.io](https://docs.monai.io/).
 - Please visit [`Project-MONAI/tutorials`](https://github.com/Project-MONAI/tutorials) for more examples, including:
-  - [`3D segmentation pipelines`](https://github.com/Project-MONAI/tutorials/tree/master/3d_segmentation),
+  - [`3D segmentation pipelines`](https://github.com/Project-MONAI/tutorials/tree/main/3d_segmentation),
   - [`Dynamic UNet`](https://github.com/Project-MONAI/tutorials/blob/main/modules/dynunet_tutorial.ipynb),
-  - [`Training acceleration`](https://github.com/Project-MONAI/tutorials/tree/master/acceleration).
+  - [`Training acceleration`](https://github.com/Project-MONAI/tutorials/tree/main/acceleration).
 
 ## Submitting to the leaderboard
 

3d_segmentation/spleen_segmentation_3d.ipynb

@@ -282,9 +282,9 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=[\"image\", \"label\"]),\n",
     "        Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        Spacingd(keys=[\"image\", \"label\"], pixdim=(\n",
     "            1.5, 1.5, 2.0), mode=(\"bilinear\", \"nearest\")),\n",
+    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        ScaleIntensityRanged(\n",
     "            keys=[\"image\"], a_min=-57, a_max=164,\n",
     "            b_min=0.0, b_max=1.0, clip=True,\n",
@@ -315,9 +315,9 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=[\"image\", \"label\"]),\n",
     "        Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        Spacingd(keys=[\"image\", \"label\"], pixdim=(\n",
     "            1.5, 1.5, 2.0), mode=(\"bilinear\", \"nearest\")),\n",
+    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        ScaleIntensityRanged(\n",
     "            keys=[\"image\"], a_min=-57, a_max=164,\n",
     "            b_min=0.0, b_max=1.0, clip=True,\n",
@@ -697,9 +697,9 @@
     "        LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        EnsureChannelFirstd(keys=[\"image\", \"label\"]),\n",
     "        Orientationd(keys=[\"image\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        Spacingd(keys=[\"image\"], pixdim=(\n",
     "            1.5, 1.5, 2.0), mode=\"bilinear\"),\n",
+    "        FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "        ScaleIntensityRanged(\n",
     "            keys=[\"image\"], a_min=-57, a_max=164,\n",
     "            b_min=0.0, b_max=1.0, clip=True,\n",
@@ -792,9 +792,9 @@
     "        LoadImaged(keys=\"image\"),\n",
     "        EnsureChannelFirstd(keys=\"image\"),\n",
     "        Orientationd(keys=[\"image\"], axcodes=\"RAS\"),\n",
-    "        FromMetaTensord(keys=\"image\"),\n",
     "        Spacingd(keys=[\"image\"], pixdim=(\n",
     "            1.5, 1.5, 2.0), mode=\"bilinear\"),\n",
+    "        FromMetaTensord(keys=\"image\"),\n",
     "        ScaleIntensityRanged(\n",
     "            keys=[\"image\"], a_min=-57, a_max=164,\n",
     "            b_min=0.0, b_max=1.0, clip=True,\n",

3d_segmentation/spleen_segmentation_3d_lightning.ipynb

@@ -272,12 +272,12 @@
     "                LoadImaged(keys=[\"image\", \"label\"]),\n",
     "                AddChanneld(keys=[\"image\", \"label\"]),\n",
     "                Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "                FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "                Spacingd(\n",
     "                    keys=[\"image\", \"label\"],\n",
     "                    pixdim=(1.5, 1.5, 2.0),\n",
     "                    mode=(\"bilinear\", \"nearest\"),\n",
     "                ),\n",
+    "                FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "                ScaleIntensityRanged(\n",
     "                    keys=[\"image\"], a_min=-57, a_max=164,\n",
     "                    b_min=0.0, b_max=1.0, clip=True,\n",
@@ -313,12 +313,12 @@
     "                LoadImaged(keys=[\"image\", \"label\"]),\n",
     "                AddChanneld(keys=[\"image\", \"label\"]),\n",
     "                Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
-    "                FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "                Spacingd(\n",
     "                    keys=[\"image\", \"label\"],\n",
     "                    pixdim=(1.5, 1.5, 2.0),\n",
     "                    mode=(\"bilinear\", \"nearest\"),\n",
     "                ),\n",
+    "                FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "                ScaleIntensityRanged(\n",
     "                    keys=[\"image\"], a_min=-57, a_max=164,\n",
     "                    b_min=0.0, b_max=1.0, clip=True,\n",

README.md

@@ -175,7 +175,7 @@ And compares the training speed and memory usage with/without AMP.
 This notebook compares the performance of `Dataset`, `CacheDataset` and `PersistentDataset`. These classes differ in how data is stored (in memory or on disk), and at which moment transforms are applied.
 #### [fast_training_tutorial](./acceleration/fast_training_tutorial.ipynb)
 This tutorial compares the training performance of pure PyTorch program and optimized program in MONAI based on NVIDIA GPU device and latest CUDA library.
-The optimization methods mainly include: `AMP`, `CacheDataset` and `Novograd`.
+The optimization methods mainly include: `AMP`, `CacheDataset`, `GPU transforms`, `ThreadDataLoader`, `DiceCELoss` and `SGD`.
 #### [multi_gpu_test](./acceleration/multi_gpu_test.ipynb)
 This notebook is a quick demo for devices, run the Ignite trainer engine on CPU, GPU and multiple GPUs.
 #### [threadbuffer_performance](./acceleration/threadbuffer_performance.ipynb)

acceleration/automatic_mixed_precision.ipynb

@@ -239,14 +239,14 @@
     "    train_transforms = Compose(\n",
     "        [\n",
     "            LoadImaged(keys=[\"image\", \"label\"]),\n",
-    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            AddChanneld(keys=[\"image\", \"label\"]),\n",
     "            Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
     "            Spacingd(\n",
     "                keys=[\"image\", \"label\"],\n",
     "                pixdim=(1.5, 1.5, 2.0),\n",
     "                mode=(\"bilinear\", \"nearest\"),\n",
     "            ),\n",
+    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            ScaleIntensityRanged(\n",
     "                keys=[\"image\"],\n",
     "                a_min=-57,\n",
@@ -283,14 +283,14 @@
     "    val_transforms = Compose(\n",
     "        [\n",
     "            LoadImaged(keys=[\"image\", \"label\"]),\n",
-    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            AddChanneld(keys=[\"image\", \"label\"]),\n",
     "            Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
     "            Spacingd(\n",
     "                keys=[\"image\", \"label\"],\n",
     "                pixdim=(1.5, 1.5, 2.0),\n",
     "                mode=(\"bilinear\", \"nearest\"),\n",
     "            ),\n",
+    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            ScaleIntensityRanged(\n",
     "                keys=[\"image\"],\n",
     "                a_min=-57,\n",

acceleration/dataset_type_performance.ipynb

@@ -398,14 +398,14 @@
     "    train_transforms = Compose(\n",
     "        [\n",
     "            LoadImaged(keys=[\"image\", \"label\"]),\n",
-    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            AddChanneld(keys=[\"image\", \"label\"]),\n",
     "            Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
     "            Spacingd(\n",
     "                keys=[\"image\", \"label\"],\n",
     "                pixdim=(1.5, 1.5, 2.0),\n",
     "                mode=(\"bilinear\", \"nearest\"),\n",
     "            ),\n",
+    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            ScaleIntensityRanged(\n",
     "                keys=[\"image\"],\n",
     "                a_min=-57,\n",
@@ -438,14 +438,14 @@
     "    val_transforms = Compose(\n",
     "        [\n",
     "            LoadImaged(keys=[\"image\", \"label\"]),\n",
-    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            AddChanneld(keys=[\"image\", \"label\"]),\n",
     "            Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
     "            Spacingd(\n",
     "                keys=[\"image\", \"label\"],\n",
     "                pixdim=(1.5, 1.5, 2.0),\n",
     "                mode=(\"bilinear\", \"nearest\"),\n",
     "            ),\n",
+    "            FromMetaTensord(keys=[\"image\", \"label\"]),\n",
     "            ScaleIntensityRanged(\n",
     "                keys=[\"image\"],\n",
     "                a_min=-57,\n",

acceleration/fast_model_training_guide.md

@@ -305,14 +305,13 @@ With all the above strategies, in this section, we introduce how to apply them t
 ### 1. Spleen segmentation
 
 - Select the algorithms based on the experiments.
-  1. As a binary segmentation task, we replaced the baseline `Dice` loss with a `DiceCE` loss, it can help improve the convergence. To achieve the target metric (mean Dice = 0.95) it reduces the number of training epochs from 200 to 50.
-  2. We tried several numerical optimizers, and finally replaced the baseline `Adam` optimizer with `Novograd`, which consistently reduce the number of training epochs from 50 to 30.
+  As a binary segmentation task, we replaced the baseline `Dice` loss with a `DiceCE` loss, it can help improve the convergence. And we tried to analyze the training curve and tuned different parameters of the network and tested several numerical optimizers, finally replaced the baseline `Adam` optimizer with `SGD`. To achieve the target metric (`mean Dice = 0.94` of the `foreground` channel only) it reduces the number of training epochs from 280 to 60.
 - Optimize GPU utilization.
   1. With `AMP`, the training speed is significantly improved and can achieve almost the same validation metric as without `AMP`.
   2. The deterministic transform results of all the spleen dataset is around 8 GB, which can be cached in a V100 GPU memory. So, we cached all the data in GPU memory and executed the following transforms in GPU directly.
 - Replace `DataLoader` with `ThreadDataLoader`. As all the data are cached in GPU, the computation of randomized transforms is on GPU and light-weighted, `ThreadDataLoader` help avoid the IPC cost of multi-processing in `DataLoader` and increase the GPU utilization.
 
-In summary, with a V100 GPU, we can achieve the training converges at a target validation mean Dice of `0.95` within one minute (`52s` on a V100 GPU, `41s` on an A100 GPU), it is approximately `200x` faster compared with the native PyTorch implementation when achieving the target metric. And each epoch is `20x` faster than the regular training.
+In summary, with a V100 GPU and the target validation `mean dice = 0.94` of the `forground` channel only,  it's more than `100x` speedup compared with the Pytorch regular implementation when achieving the same metric (validation accuracies). And every epoch is `20x` faster than regular training.
 ![spleen fast training](../figures/fast_training.png)
 
 More details are available at [Spleen fast training tutorial](https://github.com/Project-MONAI/tutorials/blob/main/acceleration/fast_training_tutorial.ipynb).