diff --git a/src/gateware/cxp_frame_pipeline.py b/src/gateware/cxp_frame_pipeline.py
index 7f9f1f9..943da60 100644
--- a/src/gateware/cxp_frame_pipeline.py
+++ b/src/gateware/cxp_frame_pipeline.py
@@ -9,6 +9,7 @@ from src.gateware.cxp_pipeline import * # for sim only
 
 from types import SimpleNamespace
 from math import lcm
+from operator import or_, add
 
 pixel_width = 16
 pixel4x_layout = [
@@ -558,7 +559,7 @@ class Pixel_Parser(Module):
             self.pixel4x.append(Record([
                 ("x", res_width),
                 ("y", res_width),
-                ("d", pixel_width),
+                ("gray", pixel_width),
                 ("stb", 1),
                 ("eof", 1), # end of frame
             ]))
@@ -588,10 +589,98 @@ class Pixel_Parser(Module):
                     pix.stb.eq(self.sink.valid[i]),
                     pix.x.eq(x_r),
                     pix.y.eq(y_r),
-                    pix.d.eq(self.sink.data[pixel_width*i:pixel_width*(i+1)]),
+                    pix.gray.eq(self.sink.data[pixel_width*i:pixel_width*(i+1)]),
                 )
             ]
 
+class ROI(Module):
+    """
+    ROI Engine. For each frame, accumulates pixels values within a
+    rectangular region of interest, and reports the total.
+    """
+    def __init__(self, pixel_4x, count_len):
+        assert count_len <= 31
+        assert len(pixel_4x) == 4
+
+        self.cfg = Record([
+            ("x0", len(pixel_4x[0].x)),
+            ("y0", len(pixel_4x[0].y)),
+            ("x1", len(pixel_4x[0].x)),
+            ("y1", len(pixel_4x[0].y)),
+        ])
+
+        self.out = Record([
+            ("update", 1),
+            ("count", count_len),
+        ])
+
+        # # #
+        #
+
+        self.roi_4x = []
+        for _ in range(4):
+            self.roi_4x.append(Record([
+                ("x_good", 1),
+                ("y_good", 1),
+                ("gray", len(pixel_4x[0].gray)),
+                ("stb", 1),
+                ("count", count_len),
+        ]))
+        
+
+        # x_good = [Signal() for _ in range(4)]
+        # y_good = [Signal() for _ in range(4)]
+        # gray = [Signal(len(pixel_4x[0].gray)) for _ in range(4)]
+        # stb = [Signal() for _ in range(4)]
+        # count = [Signal(count_len) for _ in range(4)]
+
+        for pix, roi in zip(pixel_4x, self.roi_4x):
+            self.sync += [
+                # stage 1 - generate "good" (in-ROI) signals
+                If(pix.x == self.cfg.x0,
+                    roi.x_good.eq(1)
+                ),
+                # NOTE: this gate doens't work as 4 pixes are coming in
+                If(pix.x >= self.cfg.x1,
+                    roi.x_good.eq(0)
+                ),
+
+                # This is fine because 4x pixel are on the same line
+                If(pix.y == self.cfg.y0,
+                    roi.y_good.eq(1)
+                ),
+                If(pix.y == self.cfg.y1,
+                    roi.y_good.eq(0)
+                ),
+                If(pix.eof,
+                    roi.x_good.eq(0),
+                    roi.y_good.eq(0)
+                ),
+                roi.gray.eq(pix.gray),
+                roi.stb.eq(pix.stb),
+
+                # stage 2 - accumulate
+                If((roi.stb & roi.x_good & roi.y_good),
+                    roi.count.eq(roi.count + roi.gray)
+                )
+            ]
+
+        eof = Signal()
+        # stage 3 - update
+        self.sync += [
+            eof.eq(reduce(or_, [pix.eof for pix in pixel_4x])),
+
+            self.out.update.eq(0),
+            If(eof,
+                [roi.count.eq(0) for roi in self.roi_4x],
+                self.out.update.eq(1),
+                self.out.count.eq(reduce(add, [roi.count for roi in self.roi_4x]))
+                
+            )
+        ]
+
+        
+
         
 
 class Frame_Deserializer(Module):
@@ -628,6 +717,7 @@ class Frame_Deserializer(Module):
 
         self.submodules.parser = parser = Pixel_Parser(res_width)
         self.sync += parser.y_size.eq(self.y_size)
+        self.pixel4x = parser.pixel4x
 
         mux_cases = {
             "default": [
@@ -646,8 +736,8 @@ class Frame_Deserializer(Module):
 
 
 class ROI_Pipeline(Module):
-    # TODO: but header xsize/ysize is only 24bits, change this to 24?
-    def __init__(self, res_width=32):
+    # largest x/y pixel size supported by frame header are 24 bits
+    def __init__(self, res_width=24):
 
         # NOTE: csr need to stay outside since this module need to be cdr in the CXP_FRAME_Pipeline module
         # NOTE: TapGeo other than 1X-1Y are not supported
@@ -658,6 +748,8 @@ class ROI_Pipeline(Module):
         self.submodules.header_decoder = header_decoder = Frame_Header_Decoder()
         self.submodules.deserializer = deserializer = Frame_Deserializer(res_width)
 
+        self.submodules.roi = ROI(deserializer.pixel4x, 31)
+
         self.comb += [
             deserializer.new_frame.eq(header_decoder.new_frame),
             deserializer.l_size.eq(header_decoder.metadata.l_size),