mirror of
https://github.com/huggingface/candle.git
synced 2025-06-16 02:38:10 +00:00
7b50f3e106
* More segment-anything again. * Transformer block forward. * Two-ways transformer. * Position embeddings. * Sketch the prompt encoder. * More prompt-encoder. * More prompt-encoder. * Add the main sam module. * Embed the transformer. * And hook the transformer forward step. * Build the model. * Handle the global attn indexes. * Get the model to load.
73 lines
2.2 KiB
Rust
73 lines
2.2 KiB
Rust
use candle::{DType, IndexOp, Result, Tensor, D};
|
|
use candle_nn::{layer_norm, LayerNorm, Linear, Module, VarBuilder};
|
|
|
|
use crate::model_image_encoder::ImageEncoderViT;
|
|
use crate::model_mask_decoder::MaskDecoder;
|
|
use crate::model_prompt_encoder::PromptEncoder;
|
|
|
|
#[derive(Debug)]
|
|
pub struct Sam {
|
|
image_encoder: ImageEncoderViT,
|
|
prompt_encoder: PromptEncoder,
|
|
mask_decoder: MaskDecoder,
|
|
pixel_mean: Tensor,
|
|
pixel_std: Tensor,
|
|
}
|
|
|
|
impl Sam {
|
|
pub fn new(
|
|
encoder_embed_dim: usize,
|
|
encoder_depth: usize,
|
|
encoder_num_heads: usize,
|
|
encoder_global_attn_indexes: &[usize],
|
|
vb: VarBuilder,
|
|
) -> Result<Self> {
|
|
const PROMPT_EMBED_DIM: usize = 256;
|
|
const IMAGE_SIZE: usize = 1024;
|
|
const VIT_PATCH_SIZE: usize = 16;
|
|
|
|
let image_embedding_size = IMAGE_SIZE / VIT_PATCH_SIZE;
|
|
|
|
let image_encoder = ImageEncoderViT::new(
|
|
IMAGE_SIZE,
|
|
VIT_PATCH_SIZE,
|
|
3,
|
|
encoder_embed_dim,
|
|
encoder_depth,
|
|
encoder_num_heads,
|
|
PROMPT_EMBED_DIM,
|
|
/* qkv_bias */ true,
|
|
/* use_rel_pos */ true,
|
|
/* use_abs_pos */ true,
|
|
/* window_size */ 14,
|
|
/* global_attn_indexes */ encoder_global_attn_indexes,
|
|
vb.pp("image_encoder"),
|
|
)?;
|
|
let prompt_encoder = PromptEncoder::new(
|
|
PROMPT_EMBED_DIM,
|
|
(image_embedding_size, image_embedding_size),
|
|
(IMAGE_SIZE, IMAGE_SIZE),
|
|
16,
|
|
vb.pp("prompt_encoder"),
|
|
)?;
|
|
let mask_decoder = MaskDecoder::new(
|
|
PROMPT_EMBED_DIM,
|
|
/* num_multitask_outputs */ 3,
|
|
/* iou_head_depth */ 3,
|
|
/* iou_head_hidden_dim */ 256,
|
|
vb.pp("mask_decoder"),
|
|
)?;
|
|
let pixel_mean =
|
|
Tensor::new(&[123.675f32, 116.28, 103.53], vb.device())?.reshape((3, 1, 1))?;
|
|
let pixel_std =
|
|
Tensor::new(&[58.395f32, 57.12, 57.375], vb.device())?.reshape((3, 1, 1))?;
|
|
Ok(Self {
|
|
image_encoder,
|
|
prompt_encoder,
|
|
mask_decoder,
|
|
pixel_std,
|
|
pixel_mean,
|
|
})
|
|
}
|
|
}
|