First version complete

.gitattributes

@@ -29,3 +29,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.tif filter=lfs diff=lfs merge=lfs -text

NNfunctions.py

@@ -246,20 +246,23 @@ def save_image(data, filename,cmap):
 
 def EvaluateModel(net,opt,stack):
 
+    outfile = datetime.datetime.utcnow().strftime('%H-%M-%S')
+    outfile = 'ML-SIM_%s' % outfile
+
     os.makedirs(opt.out, exist_ok=True)
 
     print(stack.shape)
     inputimg, widefield = prepimg(stack, opt)
 
     if opt.norm == 'convert' or 'minmax' in opt.norm or 'adapthist' in opt.norm:
-        cmap = 'magma'
+        cmap = 'viridis'
     else:
         cmap = 'gray'
 
     # skimage.io.imsave('%s_wf.png' % outfile,(255*widefield.numpy()).astype('uint8'))
     wf = (255*widefield.numpy()).astype('uint8')
     wf_upscaled = skimage.transform.rescale(wf,1.5,order=3,multichannel=False) # should ideally be done by drawing on client side, in javascript
-    # save_image(wf_upscaled,'%s_wf.png' % outfile,cmap)
+    save_image(wf_upscaled,'%s_wf.png' % outfile,cmap)
 
     # skimage.io.imsave('%s.tif' % outfile, inputimg.numpy())
 
@@ -276,12 +279,13 @@ def EvaluateModel(net,opt,stack):
     if opt.norm == 'convert':
         pil_sr_img = transforms.functional.rotate(pil_sr_img,-90)
 
-    #pil_sr_img.save('%s.png' % outfile) # true output for downloading, no LUT
+    # pil_sr_img.save('%s.png' % outfile) # true output for downloading, no LUT
     sr_img = np.array(pil_sr_img)
-    sr_img = exposure.equalize_adapthist(sr_img,clip_limit=0.01)
-    # skimage.io.imsave('%s.png' % outfile, sr_img) # true out for downloading, no LUT
+    # sr_img = exposure.equalize_adapthist(sr_img,clip_limit=0.01)
+    skimage.io.imsave('%s.png' % outfile, sr_img) # true out for downloading, no LUT
+
+    sr_img = skimage.transform.rescale(sr_img,1.5,order=3,multichannel=False) # should ideally be done by drawing on client side, in javascript
 
-    # sr_img = skimage.transform.rescale(sr_img,1.5,order=3,multichannel=False) # should ideally be done by drawing on client side, in javascript
-    # save_image(sr_img,'%s_sr.png' % outfile,cmap)
-    # return outfile + '_sr.png', outfile + '_wf.png', outfile + '.png'
-    return sr_img
+    save_image(sr_img,'%s_sr.png' % outfile,cmap)
+    return outfile + '_sr.png', outfile + '_wf.png', outfile + '.png'
+    # return wf, sr_img, outfile

app.py

@@ -11,49 +11,112 @@ import numpy as np
 from PIL import Image
 import io
 import base64
+import skimage
 from NNfunctions import *
 
 opt = GetOptions_allRnd_0317()
 net = LoadModel(opt)
 
-def predict(image):
-    img = np.array(image)
-    img = np.concatenate((img,img,img),axis=2)
-    img = np.transpose(img, (2,0,1))
+gr.close_all()
+
+def predict(imagefile):
+    # img = np.array(skimage.io.imread(imagefile.name))
+    # img = np.concatenate((img,img,img),axis=2)
+    # img = np.transpose(img, (2,0,1))
+
+    img = skimage.io.imread(imagefile.name)
 
     # sr,wf,out = EvaluateModel(net,opt,img,outfile)
-    sr_img = EvaluateModel(net,opt,img)
+    sr, wf, sr_download = EvaluateModel(net,opt,img)
+
+    return wf, sr, sr_download
+
+def process_example(filename):
+    basename = os.path.basename(filename)
+    basename = basename.replace('.png','.tif')
+    img = skimage.io.imread('TestImages/%s' % basename)
 
-    return sr_img
+    sr, wf, sr_download = EvaluateModel(net,opt,img)
 
+    return wf, sr
 
 title = '<h1 style="text-align: center;">ML-SIM: Reconstruction of SIM images with deep learning</h1>'
 
 description = """
-## About
-This space demonstrates the use of a semantic segmentation model to segment pets and classify them
-according to the pixels.
-## 🚀 To run
-Upload a pet image and hit submit or select one from the given examples
+This space demonstrates the ML-SIM method for reconstruction of structured illumination microscopy images.
+
+### <a href="https://opg.optica.org/boe/viewmedia.cfm?uri=boe-12-5-2720&html=true" target='_blank' > ML-SIM: universal reconstruction of structured illumination microscopy images using transfer learning </a>
+
+_Charles N. Christensen<sup>1,2,*</sup>, Edward N. Ward<sup>1</sup>, Meng Lu<sup>1</sup>, Pietro Lio<sup>2</sup>, Clemens F. Kaminski_</br></br>
+<sup>1</sup>University of Cambridge, Department of Chemical Engineering and Biotechnology, Laser Analytics Group</br>
+<sup>2</sup>University of Cambridge, Department of Computer Science and Technology, Artificial Intelligence Group</br>
+<sup> *</sup>**Author of this repository**:
+- GitHub: [charlesnchr](http://github.com/charlesnchr)
+- Email: charles.n.chr@gmail.com
+- Twitter: [charlesnchr](https://twitter.com/charlesnchr)
+
+---
+
+## 🔬 To ru ML-SIM
+Upload a TIFF image and hit submit or select one from the examples below.
+"""
+
+article = """
+
+---
+### Read more
+- <a href='https://ML-SIM.com' target='_blank'>ML-SIM.com</a>
+- <a href='https://charles-christensen.com' target='_blank'>Website</a>
+- <a href='https://github.com/charlesnchr/ML-SIM' target='_blank'>Github</a>
+- <a href='https://opg.optica.org/boe/viewmedia.cfm?uri=boe-12-5-2720&html=true' target='_blank'>Publication</a>
 """
 
-inputs = gr.inputs.Image(label="Upload a TIFF image", type = 'pil', optional=False)
+# inputs = gr.inputs.Image(label="Upload a TIFF image", type = 'pil', optional=False)
+
+inputs = gr.inputs.File(label="Upload a TIFF image", type = 'file', optional=False)
 outputs = [
-    gr.outputs.Image(label="SIM Reconstruction")
+    gr.outputs.Image(label="INPUT (Wide-field projection)"),
+    gr.outputs.Image(label="OUTPUT (ML-SIM)"),
+    gr.outputs.File(label="Download SR image" )
     # , gr.outputs.Textbox(type="auto",label="Pet Prediction")
 ]
 
-examples = [
-    "./examples/dogcat.jpeg",
-]
-
-
+examples = glob.glob('TestImages/*')
 
 interface = gr.Interface(fn=predict,
     inputs=inputs,
     outputs=outputs,
     title = title,
     description=description,
-    examples=examples
+    article=article,
+    examples=examples,
+    allow_flagging='never'
     )
 interface.launch()
+
+
+# with gr.Blocks() as interface:
+#     gr.Markdown(title)
+#     gr.Markdown(description)
+
+#     with gr.Row():
+#         input1 = gr.inputs.File(label="Upload a TIFF image", type = 'file', optional=False)
+
+#     submit_btn = gr.Button("Reconstruct")
+
+#     with gr.Row():
+#         output1 = gr.outputs.Image(label="Wide-field projection")
+#         output2 = gr.outputs.Image(label="SIM Reconstruction")
+
+#     output3 = gr.File(label="Download SR image", visible=False)
+
+#     submit_btn.click(
+#         predict,
+#         input1,
+#         [output1, output2, output3]
+#     )
+
+#     gr.Examples(examples, input1, [output1, output2, output3])
+
+
+# interface.launch()