[CityscapesSOTA] Fix incompatibility in PaddlePaddle 2.2 (PaddlePaddl…

…e#1595)
LutaoChu · Dec 29, 2021 · 81cceb3 · 81cceb3
1 parent 6800933
commit 81cceb3
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 11 deletions.
diff --git a/contrib/CityscapesSOTA/README.md b/contrib/CityscapesSOTA/README.md
@@ -8,8 +8,11 @@ Based on the above work, we made some optimizations:
 
 We achieve mIoU of **87%** on Cityscapes validation set.
 
-![demo](https://user-images.githubusercontent.com/53808988/130719591-3e0d44b4-59a8-4633-bff2-7ce7da1c52fc.gif)
+The actual effect is as follows (for high-definition pictures, please click [here](https://github.com/PaddlePaddle/PaddleSeg/blob/release/v2.0/docs/images/cityscapes.gif)).
 
+<div align="center">
+<img src=https://user-images.githubusercontent.com/30695251/144982303-51d40188-c00d-46b7-9012-41955c4e2156.gif  width = "500" />  
+</div>
 
 ## Installation
 
@@ -37,9 +40,9 @@ mkdir -p data/cityscapes
 ```
 
 Firstly please download 3 files from [Cityscapes dataset](https://www.cityscapes-dataset.com/downloads/)
-- leftImg8bit_trainvaltest.zip
-- gtFine_trainvaltest.zip
-- leftImg8bit_trainextra.zip
+- leftImg8bit_trainvaltest.zip (11GB)
+- gtFine_trainvaltest.zip (241MB)
+- leftImg8bit_trainextra.zip (44GB)
 
 Run the following commands to do the label conversion:
 ```shell
@@ -48,8 +51,13 @@ python ../../tools/convert_cityscapes.py --cityscapes_path data/cityscapes --num
 ```
 Where 'cityscapes_path' should be adjusted according to the actual dataset path. 'num_workers' determines the number of processes started and the size can be adjusted according to the actual situation.
 
-Then download Autolabelled-Data from [google drive](https://drive.google.com/file/d/1DtPo-WP-hjaOwsbj6ZxTtOo_7R_4TKRG/view?usp=sharing)
-- refinement_final_v0.zip
+Then download and uncompress Autolabelled-Data from [google drive](https://drive.google.com/file/d/1DtPo-WP-hjaOwsbj6ZxTtOo_7R_4TKRG/view?usp=sharing)
+- refinement_final_v0.zip # This file is needed for autolabelled training for recreating SOTA
+
+Delete useless `tmp` directory in `refinement_final` directory.
+```
+rm -r tmp/
+```
 
 Convert autolabelled data according to PaddleSeg data format:
 ```shell

diff --git a/contrib/CityscapesSOTA/models/mscale_ocrnet.py b/contrib/CityscapesSOTA/models/mscale_ocrnet.py
@@ -196,8 +196,5 @@ def scale_as(x, y, align_corners=False):
     '''
     y_shape = paddle.shape(y)
     x_scaled = nn.functional.interpolate(
-        x,
-        size=(y_shape[2], y_shape[3]),
-        mode='bilinear',
-        align_corners=align_corners)
+        x, size=y_shape[2:], mode='bilinear', align_corners=align_corners)
     return x_scaled
diff --git a/contrib/CityscapesSOTA/scripts/train.py b/contrib/CityscapesSOTA/scripts/train.py
@@ -207,7 +207,7 @@ def train(model,
             if (iter % save_interval == 0
                     or iter == iters) and (val_dataset is not None):
                 num_workers = 1 if num_workers > 0 else 0
-                mean_iou, acc = evaluate(
+                metrics = evaluate(
                     model,
                     val_dataset,
                     aug_eval=aug_eval,
@@ -218,6 +218,7 @@ def train(model,
                     stride=None,
                     crop_size=None,
                     num_workers=num_workers)
+                mean_iou, acc = metrics[0], metrics[1]
                 model.train()
 
             if (iter % save_interval == 0 or iter == iters) and local_rank == 0: