Modify metadata_register API for support individual metric propagation.

[yigithanyigit]

In current state of this API it can only propagate explicitly given features, In other words giving an identifier doesn't work.

For example; for propagating PSNR:
psnr_y, psnr_cb, psnr_cr should be given.

With these changes, propagating individual features and propagating with identifier possible.

[kylophone]

I'm not sure if we need an API like this yet (?). I will hold off on the reviewing this PR until we figure that out. In the meantime, I will reply to your patch in ffmpeg-devel.

[yigithanyigit]

Thanks for the comment,

I found some problems about this patch and I wanted to add for future reference.

After some tests I realized this patch doesn't work expected in some identifiers (VIF, ADM).
The main reason of the problem those identifiers has some conditional features.

For example:

vmaf/libvmaf/src/feature/integer_adm.c

Lines 2707 to 2741 in d95b69e

	if (!s->debug) return err;
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm", score, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_num", score_num, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_den", score_den, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_num_scale0", scores[0], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_den_scale0", scores[1], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_num_scale1", scores[2], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_den_scale1", scores[3], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_num_scale2", scores[4], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_den_scale2", scores[5], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_num_scale3", scores[6], index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_adm_den_scale3", scores[7], index);

vmaf/libvmaf/src/feature/integer_vif.c

Lines 692 to 746 in d95b69e

	if (!s->debug) return err;
	const double score_num =
	(double)vif.scale[0].num + (double)vif.scale[1].num +
	(double)vif.scale[2].num + (double)vif.scale[3].num;
	const double score_den =
	(double)vif.scale[0].den + (double)vif.scale[1].den +
	(double)vif.scale[2].den + (double)vif.scale[3].den;
	const double score =
	score_den == 0.0 ? 1.0f : score_num / score_den;
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif", score, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_num", score_num, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_den", score_den, index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_num_scale0", vif.scale[0].num,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_den_scale0", vif.scale[0].den,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_num_scale1", vif.scale[1].num,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_den_scale1", vif.scale[1].den,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_num_scale2", vif.scale[2].num,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_den_scale2", vif.scale[2].den,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_num_scale3", vif.scale[3].num,
	index);
	err |= vmaf_feature_collector_append_with_dict(feature_collector,
	s->feature_name_dict, "integer_vif_den_scale3", vif.scale[3].den,
	index);
	return err;

As you see those features going to append if debug enabled. However my implementation going to add those features anyway:

vmaf/libvmaf/src/feature/integer_adm.c

Lines 2757 to 2767 in d95b69e

	static const char *provided_features[] = {
	"VMAF_integer_feature_adm2_score", "integer_adm_scale0",
	"integer_adm_scale1", "integer_adm_scale2", "integer_adm_scale3",
	"integer_adm", "integer_adm_num", "integer_adm_den",
	"integer_adm_num_scale0", "integer_adm_den_scale0", "integer_adm_num_scale1",
	"integer_adm_den_scale1", "integer_adm_num_scale2", "integer_adm_den_scale2",
	"integer_adm_num_scale3", "integer_adm_den_scale3",
	NULL
	};

We should handle this somehow. Maybe we can add an option to VmafMetadataConfiguration.

Thanks

[yigithanyigit]

I significantly refactored my patch to escape thread synchronization issue

• Changed from per-feature propagation to propagating metadata when VMAF
  scores are complete
• Added frame index tracking in metadata handler to ensure ordered
  processing
• Added test coverage for non-monotonic frame processing scenarios

This change eliminates the need for complex reordering logic at the caller level and reduces potential thread synchronization issues by only propagating metadata when all necessary metrics are calculated.

[yigithanyigit]

@nilfm99 are those failures related to me? It seems like the tox tests are stuck.