add tutorial on seperating jersey number pipeline

tutorials/tutorial_seperate_jersey_number_pipeline.md

@@ -0,0 +1,96 @@
+# How can I seperate the Jersey Number Recognition pipeline from the whole pipeline to make the development faster?
+
+To do this is quite simple, you just need to make changes to 3 files overall.
+
+## Step 1
+First, you would want to change the pipeline in ```sn_gamesteate/configs/soccernet.yaml``` to only these components:
+
+```
+defaults:
+    - dataset: soccernet_gs
+    ...
+...
+pipeline:
+    - bbox_detector
+    - track
+    - jersey_number_detect
+...
+```
+
+While you're at it, you can also go ahead and change the modules for the track from ```bpbreid_strong_sort``` to ```oc_sort```.
+
+```
+defaults:
+    - dataset: soccernet_gs
+    ...
+    - modules/reid: prtreid
+    - modules/track: oc_sort
+    ...
+...
+```
+
+## Step 2
+Then you are going to change the ```sn_gamestate/jersey/mmocr_api.py``` script. 
+
+*Note: You can implement this in easyocr_api.py too*
+
+Here you are going to make same changes in 2 places. You are going to replace "jersey_number_detection" column in the detection dataframe to "jersey_number".
+
+```python
+...
+class MMOCR(DetectionLevelModule):
+    input_columns = ["bbox_ltwh"]
+    output_columns = ["jersey_number", "jersey_number_confidence"]
+    collate_fn = default_collate
+
+...
+
+    @torch.no_grad()
+    def process(self, batch, detections: pd.DataFrame, metadatas: pd.DataFrame):
+        ...
+        detections['jersey_number'] = jersey_number_detection
+        detections['jersey_number_confidence'] = jersey_number_confidence
+        ...
+...
+```
+
+## Step 3
+Finally you will need to make a minor change in the visualization script in ```sn_gamestate/visualization/pitch_visualization.py```. We do not need ```detection.role=="player"```, so we are just gonna comment it out. It is inside the "if condition" just below the line 
+```python
+#display jersey number
+```
+
+```python
+...
+class PitchVisualizationEngine(Callback):
+        ...
+            # display jersey number
+            if (
+                is_prediction
+                and self.cfg.prediction.display_jersey_number
+                and hasattr(detection, "jersey_number")
+                # and detection.role == "player"
+            ):
+            ...
+...
+```
+
+That's it then you can run the pipeline again. Note that, this will only give an output video with jersey number. 
+
+**Additional tips: If you are developing, and want to make your testing easier, please try making these changes in ```soccernet.yaml```:**
+1. Introduce ```nframes: 50``` (You can change this value to whatever you want to) under 
+```
+...
+dataset:
+    nvid: 1
+    nframes: 50
+    ...
+```
+2. Second you can run the pipeline once and save the tracker_state object for the first run. You can then load this tracker_state.pklz file under 
+```
+...
+state:
+    save_file: null
+    load_file: your_first_tracker_object.pklz
+...
+```