Fixed voc aug convert (#19)

* Fixed voc aug convert

* update getting_started.md

* add class balanced doc

.github/ISSUE_TEMPLATE/config.yml

@@ -1 +1,6 @@
 blank_issues_enabled: false
+
+contact_links:
+  - name: MMSegmentation Documentation
+    url: https://mmsegmentation.readthedocs.io
+    about: Check the docs and FAQ to see if you question is already anwsered.

docs/config.md

@@ -363,3 +363,13 @@ data = dict(
     test=dict(pipeline=test_pipeline))
 ```
 We first define the new `train_pipeline`/`test_pipeline` and pass them into `data`.
+
+Similarly, if we would like to switch from `SyncBN` to `BN` or `MMSyncBN`, we need to substitute every `norm_cfg` in the config.
+```python
+_base_ = '../pspnet/psp_r50_512x1024_40ki_cityscpaes.py'
+norm_cfg = dict(type='BN', requires_grad=True)
+model = dict(
+    backbone=dict(norm_cfg=norm_cfg),
+    decode_head=dict(norm_cfg=norm_cfg),
+    auxiliary_head=dict(norm_cfg=norm_cfg))
+```

docs/getting_started.md

@@ -132,6 +132,8 @@ Assume that you have already downloaded the checkpoints to the directory `checkp
         checkpoints/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth \
         4 --out results.pkl --eval mIoU cityscapes
     ```
+   Note: There is some inconsistency between cityscapes mIoU and our mIoU. The reason is that cityscapes average each class with class size by default.
+   We use the simple version without average for all datasets.
 
 5. Test PSPNet on cityscapes test split with 4 GPUs, and generate the png files to be submit to the official evaluation server.
 
@@ -151,8 +153,8 @@ Assume that you have already downloaded the checkpoints to the directory `checkp
         4 --format-only --options "imgfile_prefix=./pspnet_test_results"
     ```
 
-You will get png files under `./pspnet_test_results` directory.
+    You will get png files under `./pspnet_test_results` directory.
-You may run `zip -r results.zip pspnet_test_results/` and submit the zip file to [evaluation server](https://www.cityscapes-dataset.com/submit/).
+    You may run `zip -r results.zip pspnet_test_results/` and submit the zip file to [evaluation server](https://www.cityscapes-dataset.com/submit/).
 
 
 ### Image demo

docs/tutorials/training_tricks.md

@@ -26,3 +26,19 @@ model=dict(
         sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=100000)) )
 ```
 In this way, only pixels with confidence score under 0.7 are used to train. And we keep at least 100000 pixels during training.
+
+## Class Balanced Loss
+For dataset that is not balanced in classes distribution, you may change the loss weight of each class.
+Here is an example for cityscapes dataset.
+```python
+_base_ = './pspnet_r50-d8_512x1024_40k_cityscapes.py'
+model=dict(
+    decode_head=dict(
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0,
+            # DeepLab used this class weight for cityscapes
+            class_weight=[0.8373, 0.9180, 0.8660, 1.0345, 1.0166, 0.9969, 0.9754,
+                        1.0489, 0.8786, 1.0023, 0.9539, 0.9843, 1.1116, 0.9037,
+                        1.0865, 1.0955, 1.0865, 1.1529, 1.0507])))
+```
+`class_weight` will be passed into `CrossEntropyLoss` as `weight` argument. Please refer to [PyTorch Doc ](https://pytorch.org/docs/stable/nn.html?highlight=crossentropy#torch.nn.CrossEntropyLoss) for details.

tools/convert_datasets/voc_aug.py

@@ -50,8 +50,12 @@ def main():
         list(mmcv.scandir(in_dir, suffix='.mat')),
         nproc=nproc)
 
-    with open(osp.join(aug_path, 'dataset', 'trainval.txt')) as f:
+    full_aug_list = []
-        full_aug_list = [line.strip() for line in f]
+    with open(osp.join(aug_path, 'dataset', 'train.txt')) as f:
+        full_aug_list += [line.strip() for line in f]
+    with open(osp.join(aug_path, 'dataset', 'val.txt')) as f:
+        full_aug_list += [line.strip() for line in f]
+
     with open(
             osp.join(devkit_path, 'VOC2012/ImageSets/Segmentation',
                      'train.txt')) as f: