Merge pull request #24384 from Dhanwanth1803:feat-crop

Fixes #22747. Support [crop] configuration for DarkNet #24384

Request for comments. This is my first PR. 

**Merge with extra**: https://github.com/opencv/opencv_extra/pull/1112

resolves https://github.com/opencv/opencv/issues/22747

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake

modules/dnn/src/darknet/darknet_io.cpp

@@ -309,6 +309,48 @@ namespace cv {
                     fused_layer_names.push_back(last_layer);
                 }
 
+                void setCrop(int crop_height, int crop_width, int inp_height, int inp_width, bool noadjust)
+                {
+                    cv::dnn::LayerParams crop_param;
+                    crop_param.name = "CropLayer-name";
+                    std::vector<int> begin = {0, 0, (inp_height - crop_height) / 2, (inp_width - crop_width) / 2};
+                    std::vector<int> sizes = {-1, -1, crop_height, crop_width};
+                    crop_param.set("begin", DictValue::arrayInt(&begin[0], begin.size()));
+                    crop_param.set("size", DictValue::arrayInt(&sizes[0], sizes.size()));
+                    crop_param.type = "Slice";
+
+                    darknet::LayerParameter lp;
+                    std::string layer_name = cv::format("crop_%d", layer_id);
+                    lp.layer_name = layer_name;
+                    lp.layer_type = crop_param.type;
+                    lp.layerParams = crop_param;
+                    lp.bottom_indexes.push_back(last_layer);
+                    last_layer = layer_name;
+                    net->layers.push_back(lp);
+                    layer_id++;
+
+                    if (!noadjust)
+                    {
+                        cv::dnn::LayerParams params;
+                        params.set("bias_term", true);
+                        params.blobs = {
+                            Mat(1, 1, CV_32F, Scalar(2)),
+                            Mat(1, 1, CV_32F, Scalar(-1))
+                        };
+
+                        darknet::LayerParameter lp;
+                        std::string layer_name = cv::format("adjust_crop_%d", layer_id);
+                        lp.layer_name = layer_name;
+                        lp.layer_type = "Scale";
+                        lp.layerParams = params;
+                        lp.bottom_indexes.push_back(last_layer);
+                        last_layer = layer_name;
+                        net->layers.push_back(lp);
+                        layer_id++;
+                    }
+                    fused_layer_names.push_back(last_layer);
+                }
+
                 void setSoftmax()
                 {
                     cv::dnn::LayerParams softmax_param;
@@ -685,8 +727,8 @@ namespace cv {
 
                 MatShape tensor_shape(3);
                 tensor_shape[0] = net->channels;
-                tensor_shape[1] = net->width;
-                tensor_shape[2] = net->height;
+                tensor_shape[1] = net->height;
+                tensor_shape[2] = net->width;
                 net->out_channels_vec.resize(net->layers_cfg.size());
 
                 layers_counter = -1;
@@ -763,6 +805,19 @@ namespace cv {
                         tensor_shape[1] = 1;
                         tensor_shape[2] = 1;
                     }
+                    else if (layer_type == "crop")
+                    {
+                        int crop_height = getParam<int>(layer_params, "crop_height", 0);
+                        int crop_width = getParam<int>(layer_params, "crop_width", 0);
+                        bool noadjust = getParam<int>(layer_params, "noadjust", false);
+                        CV_CheckGT(crop_height, 0, "");
+                        CV_CheckGT(crop_width, 0, "");
+
+                        setParams.setCrop(crop_height, crop_width, tensor_shape[1], tensor_shape[2], noadjust);
+
+                        tensor_shape[1] = crop_height;
+                        tensor_shape[2] = crop_width;
+                    }
                     else if (layer_type == "softmax")
                     {
                         int groups = getParam<int>(layer_params, "groups", 1);
@@ -937,8 +992,8 @@ namespace cv {
 
                 MatShape tensor_shape(3);
                 tensor_shape[0] = net->channels;
-                tensor_shape[1] = net->width;
-                tensor_shape[2] = net->height;
+                tensor_shape[1] = net->height;
+                tensor_shape[2] = net->width;
                 int cv_layers_counter = -1;
                 int darknet_layers_counter = -1;
 

modules/dnn/test/test_darknet_importer.cpp

@@ -1049,6 +1049,11 @@ TEST_P(Test_Darknet_layers, avgpool_softmax)
     testDarknetLayer("avgpool_softmax");
 }
 
+TEST_P(Test_Darknet_layers, crop)
+{
+    testDarknetLayer("crop");
+}
+
 TEST_P(Test_Darknet_layers, region)
 {
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021040000)