updated DECam ROC curves and docs

docs/model_zoo.rst

@@ -30,17 +30,22 @@ DECam is a high-performance, wide-field CCD imager mounted at the prime focus of
     model_cpu = deepCR(mask='decam', device='CPU')
     model_gpu = deepCR(mask='decam', device='GPU')
 
-.. image:: https://raw.githubusercontent.com/profjsb/deepCR/master/imgs/decam.png
+.. image:: https://raw.githubusercontent.com/profjsb/deepCR/master/imgs/decam_v1.png
 
-Input images should come from calibrated images in the original unit (adu).
 The ROC curves above are produced from a test set that contains noise in cosmic ray labels.
 This causes TPR to be lower than actual because the deepCR predicted CR labels is essentially noise-free.
 
-Model is trained on g-band images but is expected to work on
-other filters as well. We are working on benchmarking on different filters
-but before that's done please proceed with caution working with other filters.
+Note 1: Output will include some bad pixels and columns, and sometimes blooming patterns.
+In particular, the blooming artifacts in the CR mask might be 1-2 pixels larger than the
+actual blooming size and cannot be excluded by subtracting by a saturation mask.
 
-For extremely short or long exposures (t_exp < 10s or t_exp > 1800s), please visually verify mask output.
+Note 2:Input images should come from calibrated images in the original unit (adu).
+
+Note 3: Model is trained on g-band images but is expected to work on
+other filters as well. We have benchmarked on g-band and r-band and are working i-band and z-band
+but only expect minor differences from the ROC curves above.
+
+Note 4: For extremely short or long exposures (t_exp < 10s or t_exp > 1800s), please visually verify mask output.
 
 Generic Ground Imaging
 ^^^^^^^^^^^^^^^^^^^^^^