Merge branch 'release-1.3.7a1' into 'release'

Release 1.3.7a1

See merge request dali/PandoraBox/pandora_plugins/plugin_mc-cnn!78

Author: duzelis

CHANGELOG.md

@@ -1,5 +1,7 @@
 # Changelog
 
+## 1.3.7a1 (June 2026)
+
 ## 1.3.6 (April 2026)
 
 ## 1.3.6a1 (March 2026)

README.md

@@ -43,9 +43,20 @@ More specifically, you can find :
 
 ### Download weights files
 
-Pretrained weights for mc-cnn fast and mc-cnn accurate neural networks are available in the weights directory :
--  mc_cnn_fast_mb_weights.pt and mc_cnn_accurate_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [[Middlebury]](#Middlebury)
--  mc_cnn_fast_data_fusion_contest.pt and mc_cnn_accurate_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [[DFC]](#DFC)
+Pretrained weights for mc-cnn fast and mc-cnn accurate neural networks are available in the MCCNN [repository](https://github.com/CNES/Pandora_MCCNN/tree/master/src/mc_cnn/weights):
+
+**Fast network**
+
+-  mc_cnn_fast_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [[Middlebury]](#Middlebury)
+-  mc_cnn_fast_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [[DFC]](#DFC)
+-  mc_cnn_fast_dw.onnx are the weights of the pretrained networks using depthwise convolutions on the Middlebury dataset [[Middlebury]](#Middlebury)
+-  mc_cnn_fast_int8_excl_01.onnx are the weights of the quantized pretrained networks on the Middlebury dataset [[Middlebury]](#Middlebury), where layers 0 and 1 are excluded from quantization
+
+**Accurate network**
+
+-  mc_cnn_accurate_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [[Middlebury]](#Middlebury)
+-  mc_cnn_accurate_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [[DFC]](#DFC)
+
 
 To download the pretrained weights:
 
@@ -54,6 +65,8 @@ wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/
 wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/mc_cnn_fast_data_fusion_contest.pt
 wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/mc_cnn_accurate_mb_weights.pt
 wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/mc_cnn_accurate_data_fusion_contest.pt
+wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/mc_cnn_fast_dw.onnx
+wget https://raw.githubusercontent.com/CNES/Pandora_MCCNN/master/mc_cnn/weights/mc_cnn_fast_int8_excl_01.onnx
 ```
 
 ### Access weights from pip package
@@ -66,7 +79,9 @@ from mc_cnn.weights import get_weights
 mc_cnn_fast_mb_weights_path = get_weights(arch="fast", training_dataset="middlebury")
 mc_cnn_fast_data_fusion_contest_path = get_weights(arch="fast", training_dataset="dfc")
 mc_cnn_accurate_mb_weights_path = get_weights(arch="accurate", training_dataset="middlebury")
-mc_cnn_accurate_data_fusion_contest = get_weights(arch="accurate", training_dataset="dfc")
+mc_cnn_accurate_data_fusion_contest_path = get_weights(arch="accurate", training_dataset="dfc")
+mc_cnn_onnx_fast_int8_path = get_weights(arch="onnx_fast_int8", training_dataset="middlebury")
+mc_cnn_onnx_fast_dw_path = get_weights(arch="onnx_fast_dw", training_dataset="middlebury")
 ```
 
 ## Output example

doc/source/getting_started.rst

@@ -37,11 +37,21 @@ More specifically, you can find :
 Pretrained Weights for MCCNN networks
 #####################################
 
-Pretrained weights for mc-cnn fast and mc-cnn accurate neural networks are available in the weights directory :
+Pretrained weights for mc-cnn fast and accurate networks are available in the MC-CNN 
+`repository <https://github.com/CNES/Pandora_MCCNN/tree/master/src/mc_cnn/weights>`_ :
 
--  mc_cnn_fast_mb_weights.pt and mc_cnn_accurate_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [Middlebury]_
+**Fast network**
+
+-  mc_cnn_fast_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [Middlebury]_
+-  mc_cnn_fast_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [DFC]_
+-  mc_cnn_fast_dw.onnx are the weights of the pretrained networks using depthwise convolutions on the Middlebury dataset [Middlebury]_
+-  mc_cnn_fast_int8_excl_01.onnx are the weights of the quantized pretrained networks on the Middlebury dataset [Middlebury]_, where layers 0 and 1 are excluded from quantization
+
+**Accurate network**
+
+-  mc_cnn_accurate_mb_weights.pt are the weights of the pretrained networks on the Middlebury dataset [Middlebury]_
+-  mc_cnn_accurate_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [DFC]_
 
--  mc_cnn_fast_data_fusion_contest.pt and mc_cnn_accurate_data_fusion_contest.pt are the weights of the pretrained networks on the Data Fusion Contest dataset [DFC]_
 
 .. [Middlebury] Scharstein, D., Hirschmüller, H., Kitajima, Y., Krathwohl, G., Nešić, N., Wang, X., & Westling, P. (2014, September). High-resolution stereo datasets with subpixel-accurate ground truth. In German conference on pattern recognition (pp. 31-42). Springer, Cham.
 

setup.cfg

@@ -41,8 +41,8 @@ setup_requires =
 # Pandora plugin MC-CNN packages dependencies
 install_requires =
     numpy
-    MCCNN==1.3.0
-    pandora==1.8.0
+    MCCNN==1.4.0a1
+    pandora==1.8.1a1
     xarray
     json-checker
     psutil

sonar-project.properties

@@ -3,7 +3,7 @@ sonar.projectKey=outilscommuns:correlateurchaine3D:pandora_plugin_mc_cnn
 
 # project metadata
 sonar.projectName=plugin_MC_CNN
-sonar.projectVersion=1.3.6
+sonar.projectVersion=1.3.7a1
 
 # path to source directories (required)
 sonar.sources=pandora_plugin_mc_cnn

tests/test_cfg_mccnn_fast.json

@@ -13,7 +13,7 @@
         "matching_cost_method": "mc_cnn",
         "subpix": 1,
         "window_size": 11,
-        "model_path": "tests/weights/mc_cnn_fast_mb_weights.pt"
+        "model_path": "tests/weights/mc_cnn_fast_int8_excl_01.onnx"
     },
     "aggregation" : {
         "aggregation_method": "cbca"

tests/test_cfg_mccnn_fast_multiband.json

@@ -14,7 +14,7 @@
         "subpix": 1,
         "window_size": 11,
         "band":"r",
-        "model_path": "tests/weights/mc_cnn_fast_mb_weights.pt"
+        "model_path": "tests/weights/mc_cnn_fast_int8_excl_01.onnx"
     },
     "aggregation" : {
         "aggregation_method": "cbca"

tests/test_plugin.py

@@ -104,7 +104,7 @@ def test_mc_cnn(self):
     def test_mc_cnn_default_values(self):
         """
         Test Pandora + plugin_mc-cnn without specifying parameters window size, subpix and model_path
-        Uses the default model path : "mc_cnn_fast_mb_weights.pt" stored in MC-CNN pip package
+        Uses the default model path : "mc_cnn_fast_int8_excl_01.onnx" stored in MC-CNN pip package
 
         """
         # Create temporary directory
@@ -122,7 +122,7 @@ def test_mc_cnn_default_values(self):
     def test_mc_cnn_multiband_values(self):
         """
         Test Pandora + plugin_mc-cnn with multiband input images
-        Uses the default model path : "mc_cnn_fast_mb_weights.pt" stored in MC-CNN pip package
+        Uses the default model path : "mc_cnn_fast_int8_excl_01.onnx" stored in MC-CNN pip package
 
         """
         # Create temporary directory
@@ -320,7 +320,7 @@ def test_invalidates_cost(self):
                 "matching_cost_method": "mc_cnn",
                 "window_size": 11,
                 "subpix": 1,
-                "model_path": "tests/weights/mc_cnn_fast_mb_weights.pt",
+                "model_path": "tests/weights/mc_cnn_fast_int8_excl_01.onnx",
             }
         )
         disparity_grids = (
@@ -525,7 +525,7 @@ def test_invalidates_cost(self):
                 "matching_cost_method": "mc_cnn",
                 "window_size": 11,
                 "subpix": 1,
-                "model_path": "tests/weights/mc_cnn_fast_mb_weights.pt",
+                "model_path": "tests/weights/mc_cnn_fast_int8_excl_01.onnx",
             }
         )
         disparity_grids = (