how to correctly close deduplication
I try close deduplication by removing the definition DRACO_ATTRIBUTE_INDICES_DEDUPLICATION_SUPPORTED and DRACO_ATTRIBUTE_VALUES_DEDUPLICATION_SUPPORTED
but I get bad compress result: I set qp - 11 qt - 10 qn -0 qg - 0 the compress seems ok when cl is between -5 to -10. when cl is set to -0 to -4 I get very big .dro file, about 106% the size of the original file
@MovingZ Are you using draco_encoder and passing negative flag values? Please note the text from the application usage:
Use negative quantization values to skip the specified attribute
@MovingZ Are you using draco_encoder and passing negative flag values? Please note the text from the application usage:
Use negative quantization values to skip the specified attribute
Already solved the problem by set deduplicate_input_values_ of obj_decoder to false. Thanks
By the way, can different attributes share the same point mapping?
I made a new decoder for the my mesh data format. Compared with LZMA, the average compress rate is 32%, but the rate of some meshes is only about 9%. I wonder how I can get a better compression.
I use SetUseBuiltInAttributeCompression(false) and use LZMA on the top of draco. It works better, but still not good enough on some meshes, compared with using LZMA directly
Can you share your model and all options that you used to compress it?
For some reason I can't share the model. But I can give the model info and data format:
IndexCount: 12786 VertexCount: 12690 VertexFormat: P3F_N4B_T2F + T4H_B4H + W4B_I4B (7 attributes: P3F - position, T2F - texcoord, others(including normal) - generic
The description of the mesh is stored in another xml file, the mesh data ONLY include mesh indices and vertices.
A: The mesh data is a bin file with raw data: indices and vertices. I use LZMA to compress it.
B: I write a new encoder to read my xml file and mesh data file. put those raw data into 7 draco attributes. Each attributes has the same indices map. Here are the options I encode with draco: -cl 6 -qp 11 -qt 10 -qn 0 -qg 11 (because my normal data is not float, I put it into generic)
After some tests, I also set SetUseBuiltInAttributeCompression(false) and use LZMA on the top of draco to improve the compression
