release-v1.1.0

README.md

@@ -91,6 +91,32 @@ The default colormap is plotly's G10, found [here](https://plotly.com/python/dis
 
 (you can google search those hex codes and get a color picker view). If you have more than 10 classes, the program uses `Light24` instead. This will give you up to 24 classes. Remember to keep your class names short, so the buttons all fit on the screen!
 
+### Example screenshots of use with example dataset
+
+#### `doodler.py`
+
+![Example 1](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py.png)
+![Example 2](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py1.png)
+![Example 3](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py2.png)
+![Example 4](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py3.png)
+![Example 5](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py4.png)
+![Example 6](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/doodler_py5.png)
+
+
+#### `refine_labels.py`
+
+![Example 1](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/refine_py.png)
+![Example 2](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/refine_py1.png)
+
+
+#### `predict_folder.py`
+
+![Example 1](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/predict_py.png)
+![Example 2](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/predict_py1.png)
+![Example 3](https://raw.githubusercontent.com/dbuscombe-usgs/dash_doodler/main/assets/logos/predict_py2.png)
+
+
+
 ### Videos
 More demonstration videos:
 
@@ -243,6 +269,20 @@ Submit a pull request through the GitHub website.
 * Each setting default now displayed on control panel
 * class_weight="balanced_subsample", min_samples_split=3
 
+03/17/21
+* max samples now 1e5, subsampled thereafter from all values (new and concatenated from file)
+* crf_refine now loops through 5 different 'rolled' versions of the label/image combo and and an average is taken. Rolling (wot I made up) is shifting an image and unary potentials on the x axis and recomputing the label raster in the shifted position, then unrolling back and averaging down the stack of unrolled label rasters
+* min_samples_split=5 in RF
+* in `predict_folder.py`, now loops through 5 different 'rolled' versions of the label/image combo and a weighted average is taken
+* added a 'sample' set to test `predict_folder.py`, and provide model
+* provided a 'clean' yml file (no version numbers) and added tqdm as a dependency
+* program now reads in default values from `src\defaults.py` or `my_defaults.py`(see below)
+* program uses SIGMA_MAX and SIGMA_MIN from `src\defaults.py` or `my_defaults.py`(see below) rather than hard-coded in
+* created `refine_labels.py` (not yet documented)
+* program now writes and reads `my_defaults.py` that keeps track of your settings preferences
+* IP address (http:https://127.0.0.1:8050/) now displayed in terminal window
+* added example workflow for sample dataset
+
 
 ## <a name="roadmap"></a>Roadmap
 

doodler.py

@@ -55,14 +55,30 @@
 ##========================================================
 DEFAULT_IMAGE_PATH = "assets/logos/dash-default.jpg"
 
-DEFAULT_PEN_WIDTH = 2 # gives line width of 2^2 = 4
-DEFAULT_CRF_DOWNSAMPLE = 2
-DEFAULT_RF_DOWNSAMPLE = 10
-DEFAULT_CRF_THETA = 40
-DEFAULT_CRF_MU = 100
-DEFAULT_MEDIAN_KERNEL = 3
-DEFAULT_RF_NESTIMATORS = 5
-DEFAULT_CRF_GTPROB = 0.9
+try:
+ from my_defaults import *
+except:
+ from defaults import *
+finally:
+ DEFAULT_PEN_WIDTH = 2
+
+ DEFAULT_CRF_DOWNSAMPLE = 2
+
+ DEFAULT_RF_DOWNSAMPLE = 10
+
+ DEFAULT_CRF_THETA = 40
+
+ DEFAULT_CRF_MU = 100
+
+ DEFAULT_MEDIAN_KERNEL = 3
+
+ DEFAULT_RF_NESTIMATORS = 3
+
+ DEFAULT_CRF_GTPROB = 0.9
+
+ SIGMA_MIN = 1
+
+ SIGMA_MAX = 16
 
 # SEG_FEATURE_TYPES = ["intensity", "edges", "texture"]
 
@@ -410,7 +426,7 @@ def download(path):
  min=1,
  max=30,
  step=1,
- value=[1, 16],
+ value=[SIGMA_MIN, SIGMA_MAX], #1, 16],
  ),
 
  html.H6(id="rf-downsample-display"),
@@ -427,8 +443,8 @@ def download(path):
  # Slider for specifying pen width
  dcc.Slider(
  id="rf-nestimators-slider",
- min=3,
- max=6,
+ min=1,
+ max=5,
  step=1,
  value=DEFAULT_RF_NESTIMATORS,
  ),
@@ -905,6 +921,26 @@ def update_output(
 
  image_list_data.append(select_image_value)
 
+ try:
+ os.remove('my_defaults.py')
+ except:
+ pass
+
+ with open('my_defaults.py', 'a') as the_file:
+ the_file.write('DEFAULT_PEN_WIDTH = {}\n'.format(pen_width))
+ the_file.write('DEFAULT_CRF_DOWNSAMPLE = {}\n'.format(crf_downsample_value))
+ the_file.write('DEFAULT_RF_DOWNSAMPLE = {}\n'.format(rf_downsample_value))
+ the_file.write('DEFAULT_CRF_THETA = {}\n'.format(crf_theta_slider_value))
+ the_file.write('DEFAULT_CRF_MU = {}\n'.format(crf_mu_slider_value))
+ the_file.write('DEFAULT_MEDIAN_KERNEL = {}\n'.format(median_filter_value))
+ the_file.write('DEFAULT_RF_NESTIMATORS = {}\n'.format(n_estimators))
+ the_file.write('DEFAULT_CRF_GTPROB = {}\n'.format(gt_prob))
+ the_file.write('SIGMA_MIN = {}\n'.format(sigma_range_slider_value[0]))
+ the_file.write('SIGMA_MAX = {}\n'.format(sigma_range_slider_value[1])) 
+
+ logging.info(datetime.now().strftime("%d-%m-%Y-%H-%M-%S"))
+ logging.info('my_defaults.py overwritten with new parameter settings')
+
 
  if len(files) == 0:
  return [
@@ -964,4 +1000,5 @@ def update_output(
 ##========================================================
 
 if __name__ == "__main__":
+ print('Go to http:https://127.0.0.1:8050/ in your web browser to use Doodler')
  app.run_server()#debug=True) #debug=True, port=8888)

install/dashdoodler-clean.yml

@@ -0,0 +1,23 @@
+name: dashdoodler
+channels:
+ - conda-forge
+ - defaults
+dependencies:
+ - python
+ - dash-core-components
+ - plotly
+ - plotly_express
+ - CairoSVG
+ - matplotlib
+ - scipy
+ - dash
+ - numpy
+ - scikit-image
+ - dash-html-components
+ - dash-table
+ - Pillow
+ - scikit-learn
+ - pandas
+ - pydensecrf
+ - cairo
+ - tqdm

install/dashdoodler.yml

@@ -20,3 +20,4 @@ dependencies:
  - pandas==1.0.3
  - pydensecrf
  - cairo
+ - tqdm

predict_folder.py

@@ -43,6 +43,7 @@
 from skimage.filters.rank import median
 from skimage.morphology import disk
 from skimage.io import imsave
+from tqdm import tqdm
 
 from tkinter import Tk
 from tkinter.filedialog import askopenfilename, askdirectory
@@ -51,22 +52,33 @@
 ## user defined parameters
 ##=========================================================
 
-multichannel = True
-intensity = True
-texture = True
-edges = True
-sigma_min=1
-sigma_max=8
-downsample_value = 2
-crf_theta_slider_value = 10
-crf_mu_slider_value = 1
-crf_downsample_factor = 1
-median_filter_value = 0
-gt_prob = 0.9
-# RF_model_file = 'RandomForestClassifier_water_land.pkl.z'
+try:
+ from my_defaults import *
+except:
+ from defaults import *
+finally:
+ DEFAULT_PEN_WIDTH = 2
+
+ DEFAULT_CRF_DOWNSAMPLE = 2
+
+ DEFAULT_RF_DOWNSAMPLE = 10
+
+ DEFAULT_CRF_THETA = 40
+
+ DEFAULT_CRF_MU = 100
+
+ DEFAULT_MEDIAN_KERNEL = 3
+
+ DEFAULT_RF_NESTIMATORS = 3
+
+ DEFAULT_CRF_GTPROB = 0.9
+
+ SIGMA_MIN = 1
+
+ SIGMA_MAX = 16
 
 Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
-RF_model_file = askopenfilename(filetypes=[("Pick classifier file","R*.z")])
+RF_model_file = askopenfilename(filetypes=[("Pick classifier file","*.z")])
 
 ##========================================================
 
@@ -95,27 +107,37 @@
 except:
  pass
 
-
+Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
 direc = askdirectory(title='Select directory of image samples', initialdir=os.getcwd())
 files = sorted(glob(direc+'/*.jpg'))
 
-
+multichannel = True
+intensity = True
+texture = True
+edges = True
 clf = load(RF_model_file) #load RF model from file
 
-for file in files:
+# DEFAULT_CRF_MU = 255
+# DEFAULT_CRF_THETA = 10
+
+for file in tqdm(files):
  print("Working on %s" % (file))
  img = img_to_ubyte_array(file) # read image into memory
 
+ Horig = img.shape[0]
+ Worig = img.shape[1]
+
  features = extract_features(
  img,
  multichannel=multichannel,
  intensity=intensity,
  edges=edges,
  texture=texture,
- sigma_min=sigma_min,
- sigma_max=sigma_max,
+ sigma_min=SIGMA_MIN,
+ sigma_max=SIGMA_MAX,
  ) # extract image features
 
+ print("Extracting features")
 
  # use model in predictive mode
  sh = features.shape
@@ -124,10 +146,50 @@
  del features
  result = result.reshape(sh[1:])
 
- result, _ = crf_refine(result, img, crf_theta_slider_value, crf_mu_slider_value, crf_downsample_factor, gt_prob) #CRF refine
+ print("CRF refinement ")
+
+ # R = []
+ # for k in np.linspace(0,img.shape[0],5):
+ # k = int(k)
+ # result2, _ = crf_refine(np.roll(result,k), np.roll(img,k), DEFAULT_CRF_THETA, DEFAULT_CRF_MU, DEFAULT_CRF_DOWNSAMPLE, DEFAULT_CRF_GTPROB) #CRF refine
+ # result2 = np.roll(result2, -k)
+ # R.append(result2)
+
+ #result = np.floor(np.mean(np.dstack(R), axis=-1)).astype('uint8')
+
+ R = []; W = []
+ counter = 0
+ for k in np.linspace(0,int(img.shape[0]/5),5):
+ k = int(k)
+ result2, _ = crf_refine(np.roll(result,k), np.roll(img,k), DEFAULT_CRF_THETA, DEFAULT_CRF_MU, DEFAULT_CRF_DOWNSAMPLE, DEFAULT_CRF_GTPROB) #CRF refine
+
+ result2 = np.roll(result2, -k)
+ R.append(result2)
+ counter +=1
+ if k==0:
+ W.append(0.1)
+ else:
+ W.append(1/np.sqrt(k))
+
+ for k in np.linspace(0,int(img.shape[0]/5),5):
+ k = int(k)
+ result2, _ = crf_refine(np.roll(result,-k), np.roll(img,-k), DEFAULT_CRF_THETA, DEFAULT_CRF_MU, DEFAULT_CRF_DOWNSAMPLE, DEFAULT_CRF_GTPROB) #CRF refine
+
+ result2 = np.roll(result2, k)
+ R.append(result2)
+ counter +=1
+ if k==0:
+ W.append(0.1)
+ else:
+ W.append(1/np.sqrt(k))
+
+ #result2 = np.floor(np.mean(np.dstack(R), axis=-1)).astype('uint8')
+ result2 = np.round(np.average(np.dstack(R), axis=-1, weights = W)).astype('uint8')
 
  # median filter
- result = median(result, disk(median_filter_value)).astype(np.uint8)
+ result = median(result, disk(DEFAULT_MEDIAN_KERNEL)).astype(np.uint8)
+
+ print("Printing to file ")
 
  imsave(file.replace(direc,results_folder).replace('.jpg','_label.png'),
  label_to_colors(result-1, img[:,:,0]==0, alpha=128, colormap=class_label_colormap, color_class_offset=0, do_alpha=False))