diff --git a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
index 03119b3fbea12cf9065e561089ca5875a8f622b0..4dfbef739e5acfb110f155320ccca4816906fc24 100644
--- a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
+++ b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
@@ -106,7 +106,6 @@ class ConvolutionInputGenerator(HLSCustomOp):
         pad = 0
         ofm_dim = compute_conv_output_dim(ifm_dim, k, stride, pad)
         assert ifm_ch % simd == 0, "SIMD must divide IFMChannels"
-        assert k % stride == 0, "stride must divide kernel size k"
         wf = int((k * k * ifm_ch) // simd)
         folded_oshape = (1, ofm_dim, ofm_dim, wf, simd)
         return folded_oshape
@@ -313,10 +312,18 @@ class ConvolutionInputGenerator(HLSCustomOp):
             "ultra": "ap_resource_uram()",
         }
         hls_ram_style = map_to_hls_ram_style[ram_style]
+
+        hls_call = node.op_type
+        # check if non optimized ConvolutionInputGenerator is needed
+        k = self.get_nodeattr("ConvKernelDim")
+        stride = self.get_nodeattr("Stride")
+        if k % stride != 0:
+            hls_call += "_kernel_stride"
+
         self.code_gen_dict["$DOCOMPUTE$"] = [
             """{}<ConvKernelDim1, IFMChannels1, Input_precision1, IFMDim1,
                 OFMDim1, SIMD1, Stride1> (in0, out, numReps, {});""".format(
-                node.op_type, hls_ram_style
+                hls_call, hls_ram_style
             )
         ]
 
diff --git a/src/finn/custom_op/im2col.py b/src/finn/custom_op/im2col.py
index 16446c15d46ee7996162f864708f7fde6cfedaf3..1ac2dad677f76b8f2aca1a04d96f4ae379940e9a 100644
--- a/src/finn/custom_op/im2col.py
+++ b/src/finn/custom_op/im2col.py
@@ -21,8 +21,6 @@ def get_im2col_indices_nchw(
     """Returns im2col indices."""
     # First figure out what the size of the output should be
     N, C, H, W = x_shape
-    assert (H + 2 * padding - field_height) % stride_y == 0
-    assert (W + 2 * padding - field_width) % stride_x == 0
     out_height = compute_conv_output_dim(H, field_height, stride_y, padding)
     out_width = compute_conv_output_dim(W, field_width, stride_x, padding)