AOMedia AV1 Codec
TxfmSearchParams Struct Reference

Defines the parameters used to perform txfm search. More...

#include <block.h>

Data Fields

int use_default_intra_tx_type
 Whether to limit the intra txfm search type to the default txfm.
 
int default_inter_tx_type_prob_thresh
 
int prune_2d_txfm_mode
 Whether to prune 2d transforms based on 1d transform results.
 
unsigned int coeff_opt_thresholds [2]
 Variable from WinnerModeParams based on current eval mode.
 
unsigned int tx_domain_dist_threshold
 Variable from WinnerModeParams based on current eval mode.
 
TX_SIZE_SEARCH_METHOD tx_size_search_method
 Variable from WinnerModeParams based on current eval mode.
 
unsigned int use_transform_domain_distortion
 Variable from WinnerModeParams based on current eval mode.
 
unsigned int skip_txfm_level
 Variable from WinnerModeParams based on current eval mode.
 
TX_MODE tx_mode_search_type
 How to search for the optimal tx_size.
 
unsigned int predict_dc_level
 
int use_qm_dist_metric
 
int mode_eval_type
 
TX_PRUNE_TYPE nn_prune_depths_for_intra_tx
 Indicates the transform depths for which RD evaluation is skipped.
 
bool enable_nn_prune_intra_tx_depths
 Indicates if NN model should be invoked to prune transform depths.
 

Detailed Description

Defines the parameters used to perform txfm search.

For the most part, this determines how various speed features are used.

Field Documentation

◆ use_default_intra_tx_type

int TxfmSearchParams::use_default_intra_tx_type

Whether to limit the intra txfm search type to the default txfm.

This could either be a result of either sequence parameter or speed features.

◆ default_inter_tx_type_prob_thresh

int TxfmSearchParams::default_inter_tx_type_prob_thresh

Probability threshold used for conditionally forcing tx type

◆ coeff_opt_thresholds

unsigned int TxfmSearchParams::coeff_opt_thresholds[2]

Variable from WinnerModeParams based on current eval mode.

See the documentation for WinnerModeParams for more detail.

◆ tx_domain_dist_threshold

unsigned int TxfmSearchParams::tx_domain_dist_threshold

Variable from WinnerModeParams based on current eval mode.

See the documentation for WinnerModeParams for more detail.

◆ tx_size_search_method

TX_SIZE_SEARCH_METHOD TxfmSearchParams::tx_size_search_method

Variable from WinnerModeParams based on current eval mode.

See the documentation for WinnerModeParams for more detail.

◆ use_transform_domain_distortion

unsigned int TxfmSearchParams::use_transform_domain_distortion

Variable from WinnerModeParams based on current eval mode.

See the documentation for WinnerModeParams for more detail.

◆ skip_txfm_level

unsigned int TxfmSearchParams::skip_txfm_level

Variable from WinnerModeParams based on current eval mode.

See the documentation for WinnerModeParams for more detail.

◆ tx_mode_search_type

TX_MODE TxfmSearchParams::tx_mode_search_type

How to search for the optimal tx_size.

If ONLY_4X4, use TX_4X4; if TX_MODE_LARGEST, use the largest tx_size for the current partition block; if TX_MODE_SELECT, search through the whole tree.

Attention
Although this looks suspicious similar to a bitstream element, this tx_mode_search_type is only used internally by the encoder, and is not written to the bitstream. It determines what kind of tx_mode would be searched. For example, we might set it to TX_MODE_LARGEST to find a good candidate, then code it as TX_MODE_SELECT.

Referenced by av1_nonrd_pick_inter_mode_sb(), and av1_nonrd_pick_intra_mode().

◆ predict_dc_level

unsigned int TxfmSearchParams::predict_dc_level

Determines whether a block can be predicted as transform skip or DC only based on residual mean and variance. Type 0 : No skip block or DC only block prediction Type 1 : Prediction of skip block based on residual mean and variance Type 2 : Prediction of skip block or DC only block based on residual mean and variance

◆ use_qm_dist_metric

int TxfmSearchParams::use_qm_dist_metric

Whether or not we should use the quantization matrix as weights for PSNR during RD search.

◆ mode_eval_type

int TxfmSearchParams::mode_eval_type

Keep track of previous mode evaluation stage type. This will be used to reset mb rd hash record when mode evaluation type changes.

◆ enable_nn_prune_intra_tx_depths

bool TxfmSearchParams::enable_nn_prune_intra_tx_depths

Indicates if NN model should be invoked to prune transform depths.

Used to signal whether NN model should be evaluated to prune the R-D evaluation of specific transform depths.


The documentation for this struct was generated from the following file: