Commit 157b353c authored by Romain Vimont's avatar Romain Vimont Committed by Thomas Daede

Do not crash on very tiny images

Commit 8412ff0b avoided some crashes on
tiny images, but it was incomplete: since it did not execute subsampled
motion estimation with 32x32 blocks when the image was smaller, it left
the array of Option<MotionVector> full of None, while
rdo_mode_decision() unwrapped the values.

If the option is None, use a default MotionVector instead to fix the
parent 44009023
......@@ -2013,7 +2013,7 @@ fn build_coarse_pmvs<T: Pixel>(fi: &FrameInvariants<T>, fs: &FrameState<T>) -> V
} else {
// the block use for motion estimation would be smaller than the whole image
vec![[Some(Default::default()); REF_FRAMES]; fi.sb_width * fi.sb_height]
vec![[None; REF_FRAMES]; fi.sb_width * fi.sb_height]
......@@ -520,7 +520,7 @@ pub fn rdo_mode_decision<T: Pixel>(
if !mv_stack.is_empty() { pmv[0] = mv_stack[0].this_mv; }
if mv_stack.len() > 1 { pmv[1] = mv_stack[1].this_mv; }
let ref_slot = ref_slot_set[i] as usize;
let cmv = pmvs[ref_slot].unwrap();
let cmv = pmvs[ref_slot].unwrap_or_else(Default::default);
let b_me = motion_estimation(fi, fs, bsize, bo, ref_frames[0], cmv, pmv);
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment