cellpose training

Author: cfusterbarcelo

README.md

@@ -94,6 +94,12 @@ models/denoising_5slices_n2v/original-denoised-difference.png
 We extract individual timepoints from the denoised TIFF stack (done manually or with Fiji).  
 We then open selected frames in the **Cellpose GUI**, annotate them, and save masks as `_seg.npy`.
 
+To annotate 3D images open cellpose GUI as:
+```
+python -m cellpose --Zstack
+```
+For more information on how to annotate images read [this docs](https://cellpose.readthedocs.io/en/latest/gui.html) and [this blog post](https://focalplane.biologists.com/2025/06/05/annotating-images-in-cellpose/).
+
 This produces training pairs like:
 
 ```

scripts/saved-2d-to-cellpose.py

@@ -0,0 +1,59 @@
+import numpy as np
+import tifffile
+from pathlib import Path
+from shutil import rmtree
+import pandas as pd
+# import ace_tools as tools
+
+# Define input/output paths
+input_dir = Path("D:/Data/Spheroids-Data-OCProject/Individual_Images/zprojection-denoised/")
+output_dir = Path("D:/Data/Spheroids-Data-OCProject/Individual_Images/zprojection-denoised-cellpose-training/")
+output_dir.mkdir(exist_ok=True)
+
+# Clean old outputs if they exist
+if output_dir.exists():
+    rmtree(output_dir)
+output_dir.mkdir()
+
+# Find all .tif volumes
+tif_files = sorted(input_dir.glob("*.tif"))
+
+# Process each volume and extract 2D labeled slices
+saved_files = []
+
+for tif_file in tif_files:
+    base_name = tif_file.stem
+    seg_file = input_dir / f"{base_name}_seg.npy"
+    
+    if not seg_file.exists():
+        continue
+    
+    volume = tifffile.imread(tif_file)
+    mask_dict = np.load(seg_file, allow_pickle=True).item()
+    mask = mask_dict["masks"]
+    
+    for z in range(mask.shape[0]):
+        if np.any(mask[z]):
+            img_slice = volume[z]
+            mask_slice = mask[z]
+
+            slice_name = f"{base_name}_z{z:02d}"
+            img_out = output_dir / f"{slice_name}.tif"
+            mask_out = output_dir / f"{slice_name}_seg.npy"
+
+            # Compose dictionary in Cellpose format
+            dummy_flows = [np.zeros_like(mask_slice, dtype=np.float32) for _ in range(3)]
+            dummy_outlines = np.zeros_like(mask_slice, dtype=np.uint8)
+
+            cellpose_mask = {
+                "masks": mask_slice.astype(np.uint16),
+                "flows": dummy_flows,
+                "outlines": dummy_outlines,
+            }
+
+            # Save image and corresponding full mask dictionary
+            tifffile.imwrite(img_out, img_slice.astype(np.uint16))
+            np.save(mask_out, cellpose_mask)
+
+            saved_files.append({"Slice Name": img_out.name, "Type": "Image"})
+            saved_files.append({"Slice Name": mask_out.name, "Type": "Mask"})

scripts/train_cellpose.bat

@@ -3,16 +3,16 @@ echo Starting Cellpose training...
 call conda activate C:\Users\Caterina\anaconda3\envs\cellpose
 
 cellpose --train ^
-  --dir D:/Data/Spheroids-Data-OCProject/Individual_Images/2d-denoised/training/ ^
+  --dir D:/Data/Spheroids-Data-OCProject/Individual_Images/zprojection-denoised-cellpose-training/ ^
   --pretrained_model nuclei ^
   --chan 0 --chan2 0 ^
   --train_size ^
   --n_epochs 50 ^
   --batch_size 8 ^
   --learning_rate 0.2 ^
   --weight_decay 1e-5 ^
-  --save_every 5 ^
-  --model_name_out nuclei_custom ^
+  --save_every 10 ^
+  --model_name_out cellpose-zprojection-denoised ^
   --mask_filter _seg.npy ^
   --verbose