@@ -4,7 +4,7 @@ experiment: "" # e.g. "_preset_medium" name experiment preceded by "_"
44output_dir : " ${pipeline.output_dir_root}/${.type}${.experiment}/${dataset.datacatalog}/${dataset.config.dataset_name}/qp${codec.tools.inner_codec.enc_configs.qp}"
55bitstream_name : " ${dataset.config.dataset_name}_qp${codec.tools.inner_codec.enc_configs.qp}"
66
7- # ============================== Configuration below are the same with fctm.yaml in FCTM =========================================
7+ # ============================== Configuration below are the same with fctm.yaml in FCTM v9 =========================================
88# NOTE: Configuration below can always be replaced with the latest fctm.yaml in FCTM if there are any incompatiable issues
99
1010verbosity : 0
@@ -24,30 +24,46 @@ enc_configs:
2424 intra_period : 1
2525 n_bit : 10
2626 parallel_encoding : False # std inner codec parallel option
27+ max_subprocess_limit : null # null = implicit CPU count limit, value = explicit (integer) limit
2728 hash_check : 0 # md5 hash check
2829
2930tools :
31+ feature_scale : # Encoder only
32+ enabled : False
33+
3034 feature_transform :
3135 enabled : False
3236
37+ feature_importance_computation :
38+ enabled : False
39+ pre_computed : null
40+
3341 feature_reduction : # reserved for restoration part as well
34- type : ' pre-trained' # by default # or 'bypass'
35- learned_model :
36- name : ' light_fedrnet' # (m71247)
42+ nn_feature_inv_transform :
43+ name : ' light_fedrnet' # (m71247) 'light_fedrnet_with_nn_codec' (m73346)
3744 selective_learning_strategy : False # by default (m71203)
3845 split_ctx : " obj" # or "seg" # or "alt1" or "dn53"
46+ model_dir : " pretrained_weights"
47+ cttc_model : True
48+
49+ channel_rearrangement :
50+ enabled : True
3951
4052 channel_removal : # (m71203)
4153 enabled : True # by default
4254 period : ${codec.enc_configs.intra_period}
43-
55+ # When nn_feature_inv_transform.selective_learning_strategy or channel_rearrangement.enabled is equal to True
56+ # channel_removal_indication should also be inferred to be True in S/W
57+ channel_restoration_indication : False # test all configs
58+ use_lcr : True # (m74096)
59+
4460 temporal_resampling_enabled : False # (m67615)
4561 # To properly enable Temporal resampling ,
4662 # temporal_resampling_enabled = True and
4763 # "cfgs/codec/vtm-23.3/encoder_lowdelay_vtm_gop4.cfg" must be used
4864
49- refinements :
50- on_restored_ftensor :
65+ refinements : # (m68664)
66+ on_restored_ftensor : # (m68391)
5167 enabled : True
5268 period : ${codec.enc_configs.intra_period}
5369 on_reduced_ftensor :
@@ -60,18 +76,33 @@ tools:
6076 resized_channel_packing : # (m71931)
6177 enabled : True
6278 period : ${codec.enc_configs.intra_period}
63-
79+ use_lcr : True # (m74096)
6480
6581 inner_codec :
6682 type : ' vtm'
6783 stash_outputs : True
6884 inner_coding_max_chroma_format : 0 # 4:0:0 only
6985 inner_coding_max_pred_constraint : 1 # Intra + uni-prediction (I, P slices)
7086 enc_configs : ${codec.enc_configs}
87+ split_ctx : ${codec.tools.feature_reduction.nn_feature_inv_transform.split_ctx}
7188 codec_paths :
7289 _root : " /path/to/vtm" # of HM, JM...
7390 enc_exe : " ${._root}/bin/EncoderAppStatic" # or TappEncoderStatic, lencod...
7491 dec_exe : " ${._root}/bin/DecoderAppStatic" # or TappDecoderStatic, ldecod...
7592 merge_exe : " ${._root}/bin/parcatStatic"
7693 cfg_file : " ${._root}/cfg/encoder_intra_vtm.cfg"
7794 default_cfg_file : " ${._root}/cfg/encoder.cfg"
95+
96+ # two variables below are needed when type is "nn_inner_codec"
97+ enc_weight : " pretrained_weights/path/to/inner_encoder_weight"
98+ dec_weight : " pretrained_weights/path/to/inner_encoder_weight"
99+
100+ sei_messages :
101+ vision_model_info :
102+ enabled : True
103+
104+ vision_model_cfg :
105+ enabled : False
106+ update : False
107+ conf_thres : 0.05
108+ max_dets : 100
0 commit comments