ncut-pytorch / backbone.py
huzey's picture
update data and model
f5de82f
raw
history blame
32.2 kB
# Author: Huzheng Yang
# %%
from typing import Optional, Tuple
from einops import rearrange
import requests
import torch
import torch.nn.functional as F
import timm
from torch import nn
import numpy as np
import os
from functools import partial
MODEL_DICT = {}
LAYER_DICT = {}
RES_DICT = {}
class SAM2(nn.Module):
def __init__(self, model_cfg='sam2_hiera_b+',):
super().__init__()
try:
from sam2.build_sam import build_sam2
except ImportError:
print("Please install segment_anything_2 from https://github.com/facebookresearch/segment-anything-2.git")
return
config_dict = {
'sam2_hiera_l': ("sam2_hiera_large.pt", "https://dl.fbaipublicfiles.com/segment_anything_2/072824/sam2_hiera_large.pt"),
'sam2_hiera_b+': ("sam2_hiera_base_plus.pt", "https://dl.fbaipublicfiles.com/segment_anything_2/072824/sam2_hiera_base_plus.pt"),
'sam2_hiera_s': ("sam2_hiera_small.pt", "https://dl.fbaipublicfiles.com/segment_anything_2/072824/sam2_hiera_small.pt"),
'sam2_hiera_t': ("sam2_hiera_tiny.pt", "https://dl.fbaipublicfiles.com/segment_anything_2/072824/sam2_hiera_tiny.pt"),
}
filename, url = config_dict[model_cfg]
if not os.path.exists(filename):
print(f"Downloading {url}")
r = requests.get(url)
with open(filename, 'wb') as f:
f.write(r.content)
sam2_checkpoint = filename
device = 'cuda' if torch.cuda.is_available() else 'cpu'
sam2_model = build_sam2(model_cfg, sam2_checkpoint, device=device)
image_encoder = sam2_model.image_encoder
image_encoder.eval()
from sam2.modeling.backbones.hieradet import do_pool
from sam2.modeling.backbones.utils import window_partition, window_unpartition
def new_forward(self, x: torch.Tensor) -> torch.Tensor:
shortcut = x # B, H, W, C
x = self.norm1(x)
# Skip connection
if self.dim != self.dim_out:
shortcut = do_pool(self.proj(x), self.pool)
# Window partition
window_size = self.window_size
if window_size > 0:
H, W = x.shape[1], x.shape[2]
x, pad_hw = window_partition(x, window_size)
# Window Attention + Q Pooling (if stage change)
x = self.attn(x)
if self.q_stride:
# Shapes have changed due to Q pooling
window_size = self.window_size // self.q_stride[0]
H, W = shortcut.shape[1:3]
pad_h = (window_size - H % window_size) % window_size
pad_w = (window_size - W % window_size) % window_size
pad_hw = (H + pad_h, W + pad_w)
# Reverse window partition
if self.window_size > 0:
x = window_unpartition(x, window_size, pad_hw, (H, W))
self.attn_output = x.clone()
x = shortcut + self.drop_path(x)
# MLP
mlp_out = self.mlp(self.norm2(x))
self.mlp_output = mlp_out.clone()
x = x + self.drop_path(mlp_out)
self.block_output = x.clone()
return x
setattr(image_encoder.trunk.blocks[0].__class__, 'forward', new_forward)
self.image_encoder = image_encoder
@torch.no_grad()
def forward(self, x: torch.Tensor) -> torch.Tensor:
output = self.image_encoder(x)
attn_outputs, mlp_outputs, block_outputs = [], [], []
for block in self.image_encoder.trunk.blocks:
attn_outputs.append(block.attn_output)
mlp_outputs.append(block.mlp_output)
block_outputs.append(block.block_output)
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["SAM2(sam2_hiera_t)"] = partial(SAM2, model_cfg='sam2_hiera_t')
LAYER_DICT["SAM2(sam2_hiera_t)"] = 12
RES_DICT["SAM2(sam2_hiera_t)"] = (1024, 1024)
MODEL_DICT["SAM2(sam2_hiera_s)"] = partial(SAM2, model_cfg='sam2_hiera_s')
LAYER_DICT["SAM2(sam2_hiera_s)"] = 16
RES_DICT["SAM2(sam2_hiera_s)"] = (1024, 1024)
MODEL_DICT["SAM2(sam2_hiera_b+)"] = partial(SAM2, model_cfg='sam2_hiera_b+')
LAYER_DICT["SAM2(sam2_hiera_b+)"] = 24
RES_DICT["SAM2(sam2_hiera_b+)"] = (1024, 1024)
MODEL_DICT["SAM2(sam2_hiera_l)"] = partial(SAM2, model_cfg='sam2_hiera_l')
LAYER_DICT["SAM2(sam2_hiera_l)"] = 48
RES_DICT["SAM2(sam2_hiera_l)"] = (1024, 1024)
class SAM(torch.nn.Module):
def __init__(self, **kwargs):
super().__init__(**kwargs)
from segment_anything import sam_model_registry, SamPredictor
from segment_anything.modeling.sam import Sam
checkpoint = "sam_vit_b_01ec64.pth"
if not os.path.exists(checkpoint):
checkpoint_url = (
"https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth"
)
import requests
r = requests.get(checkpoint_url)
with open(checkpoint, "wb") as f:
f.write(r.content)
sam: Sam = sam_model_registry["vit_b"](checkpoint=checkpoint)
from segment_anything.modeling.image_encoder import (
window_partition,
window_unpartition,
)
def new_block_forward(self, x: torch.Tensor) -> torch.Tensor:
shortcut = x
x = self.norm1(x)
# Window partition
if self.window_size > 0:
H, W = x.shape[1], x.shape[2]
x, pad_hw = window_partition(x, self.window_size)
x = self.attn(x)
# Reverse window partition
if self.window_size > 0:
x = window_unpartition(x, self.window_size, pad_hw, (H, W))
self.attn_output = x.clone()
x = shortcut + x
mlp_outout = self.mlp(self.norm2(x))
self.mlp_output = mlp_outout.clone()
x = x + mlp_outout
self.block_output = x.clone()
return x
setattr(sam.image_encoder.blocks[0].__class__, "forward", new_block_forward)
self.image_encoder = sam.image_encoder
self.image_encoder.eval()
@torch.no_grad()
def forward(self, x: torch.Tensor) -> torch.Tensor:
with torch.no_grad():
x = torch.nn.functional.interpolate(x, size=(1024, 1024), mode="bilinear")
out = self.image_encoder(x)
attn_outputs, mlp_outputs, block_outputs = [], [], []
for i, blk in enumerate(self.image_encoder.blocks):
attn_outputs.append(blk.attn_output)
mlp_outputs.append(blk.mlp_output)
block_outputs.append(blk.block_output)
attn_outputs = torch.stack(attn_outputs)
mlp_outputs = torch.stack(mlp_outputs)
block_outputs = torch.stack(block_outputs)
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["SAM(sam_vit_b)"] = partial(SAM)
LAYER_DICT["SAM(sam_vit_b)"] = 12
RES_DICT["SAM(sam_vit_b)"] = (1024, 1024)
class MobileSAM(nn.Module):
def __init__(self, **kwargs):
super().__init__(**kwargs)
from mobile_sam import sam_model_registry
url = "https://raw.githubusercontent.com/ChaoningZhang/MobileSAM/master/weights/mobile_sam.pt"
model_type = "vit_t"
sam_checkpoint = "mobile_sam.pt"
if not os.path.exists(sam_checkpoint):
import requests
r = requests.get(url)
with open(sam_checkpoint, "wb") as f:
f.write(r.content)
mobile_sam = sam_model_registry[model_type](checkpoint=sam_checkpoint)
def new_forward_fn(self, x):
shortcut = x
x = self.conv1(x)
x = self.act1(x)
x = self.conv2(x)
x = self.act2(x)
self.attn_output = rearrange(x.clone(), "b c h w -> b h w c")
x = self.conv3(x)
self.mlp_output = rearrange(x.clone(), "b c h w -> b h w c")
x = self.drop_path(x)
x += shortcut
x = self.act3(x)
self.block_output = rearrange(x.clone(), "b c h w -> b h w c")
return x
setattr(
mobile_sam.image_encoder.layers[0].blocks[0].__class__,
"forward",
new_forward_fn,
)
def new_forward_fn2(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
res_x = x
if H == self.window_size and W == self.window_size:
x = self.attn(x)
else:
x = x.view(B, H, W, C)
pad_b = (self.window_size - H % self.window_size) % self.window_size
pad_r = (self.window_size - W % self.window_size) % self.window_size
padding = pad_b > 0 or pad_r > 0
if padding:
x = F.pad(x, (0, 0, 0, pad_r, 0, pad_b))
pH, pW = H + pad_b, W + pad_r
nH = pH // self.window_size
nW = pW // self.window_size
# window partition
x = (
x.view(B, nH, self.window_size, nW, self.window_size, C)
.transpose(2, 3)
.reshape(B * nH * nW, self.window_size * self.window_size, C)
)
x = self.attn(x)
# window reverse
x = (
x.view(B, nH, nW, self.window_size, self.window_size, C)
.transpose(2, 3)
.reshape(B, pH, pW, C)
)
if padding:
x = x[:, :H, :W].contiguous()
x = x.view(B, L, C)
hw = np.sqrt(x.shape[1]).astype(int)
self.attn_output = rearrange(x.clone(), "b (h w) c -> b h w c", h=hw)
x = res_x + self.drop_path(x)
x = x.transpose(1, 2).reshape(B, C, H, W)
x = self.local_conv(x)
x = x.view(B, C, L).transpose(1, 2)
mlp_output = self.mlp(x)
self.mlp_output = rearrange(
mlp_output.clone(), "b (h w) c -> b h w c", h=hw
)
x = x + self.drop_path(mlp_output)
self.block_output = rearrange(x.clone(), "b (h w) c -> b h w c", h=hw)
return x
setattr(
mobile_sam.image_encoder.layers[1].blocks[0].__class__,
"forward",
new_forward_fn2,
)
mobile_sam.eval()
self.image_encoder = mobile_sam.image_encoder
@torch.no_grad()
def forward(self, x):
with torch.no_grad():
x = torch.nn.functional.interpolate(x, size=(1024, 1024), mode="bilinear")
out = self.image_encoder(x)
attn_outputs, mlp_outputs, block_outputs = [], [], []
for i_layer in range(len(self.image_encoder.layers)):
for i_block in range(len(self.image_encoder.layers[i_layer].blocks)):
blk = self.image_encoder.layers[i_layer].blocks[i_block]
attn_outputs.append(blk.attn_output)
mlp_outputs.append(blk.mlp_output)
block_outputs.append(blk.block_output)
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["MobileSAM"] = partial(MobileSAM)
LAYER_DICT["MobileSAM"] = 12
RES_DICT["MobileSAM"] = (1024, 1024)
class DiNOv2(torch.nn.Module):
def __init__(self, ver="dinov2_vitb14_reg", num_reg=5):
super().__init__()
self.dinov2 = torch.hub.load("facebookresearch/dinov2", ver)
self.dinov2.requires_grad_(False)
self.dinov2.eval()
self.num_reg = num_reg
def new_block_forward(self, x: torch.Tensor) -> torch.Tensor:
def attn_residual_func(x):
return self.ls1(self.attn(self.norm1(x)))
def ffn_residual_func(x):
return self.ls2(self.mlp(self.norm2(x)))
attn_output = attn_residual_func(x)
hw = np.sqrt(attn_output.shape[1] - num_reg).astype(int)
self.attn_output = rearrange(
attn_output.clone()[:, num_reg:], "b (h w) c -> b h w c", h=hw
)
x = x + attn_output
mlp_output = ffn_residual_func(x)
self.mlp_output = rearrange(
mlp_output.clone()[:, num_reg:], "b (h w) c -> b h w c", h=hw
)
x = x + mlp_output
block_output = x
self.block_output = rearrange(
block_output.clone()[:, num_reg:], "b (h w) c -> b h w c", h=hw
)
return x
setattr(self.dinov2.blocks[0].__class__, "forward", new_block_forward)
@torch.no_grad()
def forward(self, x):
out = self.dinov2(x)
attn_outputs, mlp_outputs, block_outputs = [], [], []
for i, blk in enumerate(self.dinov2.blocks):
attn_outputs.append(blk.attn_output)
mlp_outputs.append(blk.mlp_output)
block_outputs.append(blk.block_output)
attn_outputs = torch.stack(attn_outputs)
mlp_outputs = torch.stack(mlp_outputs)
block_outputs = torch.stack(block_outputs)
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["DiNOv2reg(dinov2_vitb14_reg)"] = partial(DiNOv2, ver="dinov2_vitb14_reg", num_reg=5)
LAYER_DICT["DiNOv2reg(dinov2_vitb14_reg)"] = 12
RES_DICT["DiNOv2reg(dinov2_vitb14_reg)"] = (672, 672)
MODEL_DICT["DiNOv2(dinov2_vitb14)"] = partial(DiNOv2, ver="dinov2_vitb14", num_reg=1)
LAYER_DICT["DiNOv2(dinov2_vitb14)"] = 12
RES_DICT["DiNOv2(dinov2_vitb14)"] = (672, 672)
class DiNO(nn.Module):
def __init__(self, ver="dino_vitb8"):
super().__init__()
model = torch.hub.load('facebookresearch/dino:main', ver)
model = model.eval()
def remove_cls_and_reshape(x):
x = x.clone()
x = x[:, 1:]
hw = np.sqrt(x.shape[1]).astype(int)
x = rearrange(x, "b (h w) c -> b h w c", h=hw)
return x
def new_forward(self, x, return_attention=False):
y, attn = self.attn(self.norm1(x))
self.attn_output = remove_cls_and_reshape(y.clone())
if return_attention:
return attn
x = x + self.drop_path(y)
mlp_output = self.mlp(self.norm2(x))
self.mlp_output = remove_cls_and_reshape(mlp_output.clone())
x = x + self.drop_path(mlp_output)
self.block_output = remove_cls_and_reshape(x.clone())
return x
setattr(model.blocks[0].__class__, "forward", new_forward)
self.model = model
self.model.eval()
self.model.requires_grad_(False)
def forward(self, x):
out = self.model(x)
attn_outputs = [block.attn_output for block in self.model.blocks]
mlp_outputs = [block.mlp_output for block in self.model.blocks]
block_outputs = [block.block_output for block in self.model.blocks]
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["DiNO(dino_vitb8)"] = partial(DiNO)
LAYER_DICT["DiNO(dino_vitb8)"] = 12
RES_DICT["DiNO(dino_vitb8)"] = (448, 448)
def resample_position_embeddings(embeddings, h, w):
cls_embeddings = embeddings[0]
patch_embeddings = embeddings[1:] # [14*14, 768]
hw = np.sqrt(patch_embeddings.shape[0]).astype(int)
patch_embeddings = rearrange(patch_embeddings, "(h w) c -> c h w", h=hw)
patch_embeddings = F.interpolate(patch_embeddings.unsqueeze(0), size=(h, w), mode="nearest").squeeze(0)
patch_embeddings = rearrange(patch_embeddings, "c h w -> (h w) c")
embeddings = torch.cat([cls_embeddings.unsqueeze(0), patch_embeddings], dim=0)
return embeddings
# class CLIP(torch.nn.Module):
# def __init__(self, ver="openai/clip-vit-base-patch16"):
# super().__init__()
# from transformers import CLIPProcessor, CLIPModel
# model = CLIPModel.from_pretrained(ver)
# # resample the patch embeddings to 56x56, take 896x896 input
# embeddings = model.vision_model.embeddings.position_embedding.weight
# embeddings = resample_position_embeddings(embeddings, 42, 42)
# model.vision_model.embeddings.position_embedding.weight = nn.Parameter(embeddings)
# model.vision_model.embeddings.position_ids = torch.arange(0, 1+56*56)
# # processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
# self.model = model.eval()
# def new_forward(
# self,
# hidden_states: torch.Tensor,
# attention_mask: torch.Tensor,
# causal_attention_mask: torch.Tensor,
# output_attentions: Optional[bool] = False,
# ) -> Tuple[torch.FloatTensor]:
# residual = hidden_states
# hidden_states = self.layer_norm1(hidden_states)
# hidden_states, attn_weights = self.self_attn(
# hidden_states=hidden_states,
# attention_mask=attention_mask,
# causal_attention_mask=causal_attention_mask,
# output_attentions=output_attentions,
# )
# hw = np.sqrt(hidden_states.shape[1] - 1).astype(int)
# self.attn_output = rearrange(
# hidden_states.clone()[:, 1:], "b (h w) c -> b h w c", h=hw
# )
# hidden_states = residual + hidden_states
# residual = hidden_states
# hidden_states = self.layer_norm2(hidden_states)
# hidden_states = self.mlp(hidden_states)
# self.mlp_output = rearrange(
# hidden_states.clone()[:, 1:], "b (h w) c -> b h w c", h=hw
# )
# hidden_states = residual + hidden_states
# outputs = (hidden_states,)
# if output_attentions:
# outputs += (attn_weights,)
# self.block_output = rearrange(
# hidden_states.clone()[:, 1:], "b (h w) c -> b h w c", h=hw
# )
# return outputs
# setattr(
# self.model.vision_model.encoder.layers[0].__class__, "forward", new_forward
# )
# @torch.no_grad()
# def forward(self, x):
# out = self.model.vision_model(x)
# attn_outputs, mlp_outputs, block_outputs = [], [], []
# for i, blk in enumerate(self.model.vision_model.encoder.layers):
# attn_outputs.append(blk.attn_output)
# mlp_outputs.append(blk.mlp_output)
# block_outputs.append(blk.block_output)
# attn_outputs = torch.stack(attn_outputs)
# mlp_outputs = torch.stack(mlp_outputs)
# block_outputs = torch.stack(block_outputs)
# return attn_outputs, mlp_outputs, block_outputs
# MODEL_DICT["CLIP(openai/clip-vit-base-patch16)"] = partial(CLIP, ver="openai/clip-vit-base-patch16")
# LAYER_DICT["CLIP(openai/clip-vit-base-patch16)"] = 12
# RES_DICT["CLIP(openai/clip-vit-base-patch16)"] = (896, 896)
class OpenCLIPViT(nn.Module):
def __init__(self, version='ViT-B-16', pretrained='laion2b_s34b_b88k'):
super().__init__()
try:
import open_clip
except ImportError:
print("Please install open_clip to use this class.")
return
model, _, _ = open_clip.create_model_and_transforms(version, pretrained=pretrained)
positional_embedding = resample_position_embeddings(model.visual.positional_embedding, 42, 42)
model.visual.positional_embedding = nn.Parameter(positional_embedding)
def new_forward(
self,
q_x: torch.Tensor,
k_x: Optional[torch.Tensor] = None,
v_x: Optional[torch.Tensor] = None,
attn_mask: Optional[torch.Tensor] = None,
):
def remove_cls_and_reshape(x):
x = x.clone()
x = x[1:]
hw = np.sqrt(x.shape[0]).astype(int)
x = rearrange(x, "(h w) b c -> b h w c", h=hw)
return x
k_x = self.ln_1_kv(k_x) if hasattr(self, "ln_1_kv") and k_x is not None else None
v_x = self.ln_1_kv(v_x) if hasattr(self, "ln_1_kv") and v_x is not None else None
attn_output = self.attention(q_x=self.ln_1(q_x), k_x=k_x, v_x=v_x, attn_mask=attn_mask)
self.attn_output = remove_cls_and_reshape(attn_output.clone())
x = q_x + self.ls_1(attn_output)
mlp_output = self.mlp(self.ln_2(x))
self.mlp_output = remove_cls_and_reshape(mlp_output.clone())
x = x + self.ls_2(mlp_output)
self.block_output = remove_cls_and_reshape(x.clone())
return x
setattr(model.visual.transformer.resblocks[0].__class__, "forward", new_forward)
self.model = model
self.model.eval()
def forward(self, x):
out = self.model(x)
attn_outputs, mlp_outputs, block_outputs = [], [], []
for block in self.model.visual.transformer.resblocks:
attn_outputs.append(block.attn_output)
mlp_outputs.append(block.mlp_output)
block_outputs.append(block.block_output)
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["CLIP(ViT-B-16/openai)"] = partial(OpenCLIPViT, version='ViT-B-16', pretrained='openai')
LAYER_DICT["CLIP(ViT-B-16/openai)"] = 12
RES_DICT["CLIP(ViT-B-16/openai)"] = (672, 672)
MODEL_DICT["CLIP(ViT-B-16/laion2b_s34b_b88k)"] = partial(OpenCLIPViT, version='ViT-B-16', pretrained='laion2b_s34b_b88k')
LAYER_DICT["CLIP(ViT-B-16/laion2b_s34b_b88k)"] = 12
RES_DICT["CLIP(ViT-B-16/laion2b_s34b_b88k)"] = (672, 672)
class EVA02(nn.Module):
def __init__(self, **kwargs):
super().__init__(**kwargs)
model = timm.create_model(
'eva02_base_patch14_448.mim_in22k_ft_in1k',
pretrained=True,
num_classes=0, # remove classifier nn.Linear
)
model = model.eval()
def new_forward(self, x, rope: Optional[torch.Tensor] = None, attn_mask: Optional[torch.Tensor] = None):
def remove_cls_and_reshape(x):
x = x.clone()
x = x[:, 1:]
hw = np.sqrt(x.shape[1]).astype(int)
x = rearrange(x, "b (h w) c -> b h w c", h=hw)
return x
if self.gamma_1 is None:
attn_output = self.attn(self.norm1(x), rope=rope, attn_mask=attn_mask)
self.attn_output = remove_cls_and_reshape(attn_output.clone())
x = x + self.drop_path1(attn_output)
mlp_output = self.mlp(self.norm2(x))
self.mlp_output = remove_cls_and_reshape(mlp_output.clone())
x = x + self.drop_path2(mlp_output)
else:
attn_output = self.attn(self.norm1(x), rope=rope, attn_mask=attn_mask)
self.attn_output = remove_cls_and_reshape(attn_output.clone())
x = x + self.drop_path1(self.gamma_1 * attn_output)
mlp_output = self.mlp(self.norm2(x))
self.mlp_output = remove_cls_and_reshape(mlp_output.clone())
x = x + self.drop_path2(self.gamma_2 * mlp_output)
self.block_output = remove_cls_and_reshape(x.clone())
return x
setattr(model.blocks[0].__class__, "forward", new_forward)
self.model = model
def forward(self, x):
out = self.model(x)
attn_outputs = [block.attn_output for block in self.model.blocks]
mlp_outputs = [block.mlp_output for block in self.model.blocks]
block_outputs = [block.block_output for block in self.model.blocks]
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["CLIP(eva02_base_patch14_448.mim_in22k_ft_in1k)"] = partial(EVA02)
LAYER_DICT["CLIP(eva02_base_patch14_448.mim_in22k_ft_in1k)"] = 12
RES_DICT["CLIP(eva02_base_patch14_448.mim_in22k_ft_in1k)"] = (448, 448)
class CLIPConvnext(nn.Module):
def __init__(self):
super().__init__()
try:
import open_clip
except ImportError:
print("Please install open_clip to use this class.")
return
model, _, _ = open_clip.create_model_and_transforms('convnext_base_w_320', pretrained='laion_aesthetic_s13b_b82k')
def new_forward(self, x):
shortcut = x
x = self.conv_dw(x)
if self.use_conv_mlp:
x = self.norm(x)
x = self.mlp(x)
else:
x = x.permute(0, 2, 3, 1)
x = self.norm(x)
x = self.mlp(x)
x = x.permute(0, 3, 1, 2)
if self.gamma is not None:
x = x.mul(self.gamma.reshape(1, -1, 1, 1))
x = self.drop_path(x) + self.shortcut(shortcut)
self.block_output = rearrange(x.clone(), "b c h w -> b h w c")
return x
setattr(model.visual.trunk.stages[0].blocks[0].__class__, "forward", new_forward)
self.model = model
self.model.eval()
def forward(self, x):
out = self.model(x)
block_outputs = []
for stage in self.model.visual.trunk.stages:
for block in stage.blocks:
block_outputs.append(block.block_output)
return {
'attn': None,
'mlp': None,
'block': block_outputs
}
MODEL_DICT["CLIP(convnext_base_w_320/laion_aesthetic_s13b_b82k)"] = partial(CLIPConvnext)
LAYER_DICT["CLIP(convnext_base_w_320/laion_aesthetic_s13b_b82k)"] = 36
RES_DICT["CLIP(convnext_base_w_320/laion_aesthetic_s13b_b82k)"] = (960, 960)
class MAE(timm.models.vision_transformer.VisionTransformer):
def __init__(self, **kwargs):
super(MAE, self).__init__(**kwargs)
sd = torch.hub.load_state_dict_from_url(
"https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_base.pth"
)
checkpoint_model = sd["model"]
state_dict = self.state_dict()
for k in ["head.weight", "head.bias"]:
if (
k in checkpoint_model
and checkpoint_model[k].shape != state_dict[k].shape
):
print(f"Removing key {k} from pretrained checkpoint")
del checkpoint_model[k]
# load pre-trained model
msg = self.load_state_dict(checkpoint_model, strict=False)
print(msg)
# resample the patch embeddings to 56x56, take 896x896 input
pos_embed = self.pos_embed[0]
pos_embed = resample_position_embeddings(pos_embed, 42, 42)
self.pos_embed = nn.Parameter(pos_embed.unsqueeze(0))
self.img_size = (672, 672)
self.patch_embed.img_size = (672, 672)
self.requires_grad_(False)
self.eval()
def forward(self, x):
self.saved_attn_node = self.ls1(self.attn(self.norm1(x)))
x = x + self.saved_attn_node.clone()
self.saved_mlp_node = self.ls2(self.mlp(self.norm2(x)))
x = x + self.saved_mlp_node.clone()
self.saved_block_output = x.clone()
return x
setattr(self.blocks[0].__class__, "forward", forward)
def forward(self, x):
out = super().forward(x)
def remove_cls_and_reshape(x):
x = x.clone()
x = x[:, 1:]
hw = np.sqrt(x.shape[1]).astype(int)
x = rearrange(x, "b (h w) c -> b h w c", h=hw)
return x
attn_outputs = [remove_cls_and_reshape(block.saved_attn_node) for block in self.blocks]
mlp_outputs = [remove_cls_and_reshape(block.saved_mlp_node) for block in self.blocks]
block_outputs = [remove_cls_and_reshape(block.saved_block_output) for block in self.blocks]
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["MAE(vit_base)"] = partial(MAE)
LAYER_DICT["MAE(vit_base)"] = 12
RES_DICT["MAE(vit_base)"] = (672, 672)
class ImageNet(nn.Module):
def __init__(self, **kwargs):
super().__init__(**kwargs)
model = timm.create_model(
'vit_base_patch16_224.augreg2_in21k_ft_in1k',
pretrained=True,
num_classes=0, # remove classifier nn.Linear
)
# resample the patch embeddings to 56x56, take 896x896 input
pos_embed = model.pos_embed[0]
pos_embed = resample_position_embeddings(pos_embed, 42, 42)
model.pos_embed = nn.Parameter(pos_embed.unsqueeze(0))
model.img_size = (672, 672)
model.patch_embed.img_size = (672, 672)
model.requires_grad_(False)
model.eval()
def forward(self, x):
self.saved_attn_node = self.ls1(self.attn(self.norm1(x)))
x = x + self.saved_attn_node.clone()
self.saved_mlp_node = self.ls2(self.mlp(self.norm2(x)))
x = x + self.saved_mlp_node.clone()
self.saved_block_output = x.clone()
return x
setattr(model.blocks[0].__class__, "forward", forward)
self.model = model
def forward(self, x):
out = self.model(x)
def remove_cls_and_reshape(x):
x = x.clone()
x = x[:, 1:]
hw = np.sqrt(x.shape[1]).astype(int)
x = rearrange(x, "b (h w) c -> b h w c", h=hw)
return x
attn_outputs = [remove_cls_and_reshape(block.saved_attn_node) for block in self.model.blocks]
mlp_outputs = [remove_cls_and_reshape(block.saved_mlp_node) for block in self.model.blocks]
block_outputs = [remove_cls_and_reshape(block.saved_block_output) for block in self.model.blocks]
return {
'attn': attn_outputs,
'mlp': mlp_outputs,
'block': block_outputs
}
MODEL_DICT["ImageNet(vit_base)"] = partial(ImageNet)
LAYER_DICT["ImageNet(vit_base)"] = 12
RES_DICT["ImageNet(vit_base)"] = (672, 672)
def download_all_models():
for model_name in MODEL_DICT:
print(f"Downloading {model_name}")
model = MODEL_DICT[model_name]()
def get_all_model_names():
return list(MODEL_DICT.keys())
def get_model(model_name):
return MODEL_DICT[model_name]()
@torch.no_grad()
def extract_features(images, model, model_name, node_type, layer, batch_size=8):
use_cuda = torch.cuda.is_available()
if use_cuda:
model = model.cuda()
chunked_idxs = torch.split(torch.arange(images.shape[0]), batch_size)
outputs = []
for idxs in chunked_idxs:
inp = images[idxs]
if use_cuda:
inp = inp.cuda()
out = model(inp) # {'attn': [B, H, W, C], 'mlp': [B, H, W, C], 'block': [B, H, W, C]}
out = out[node_type]
if out is None:
raise ValueError(f"Node type {node_type} not found in model {model_name}")
out = out[layer]
# normalize
out = F.normalize(out, dim=-1)
outputs.append(out.cpu().float())
outputs = torch.cat(outputs, dim=0)
return outputs
if __name__ == '__main__':
inp = torch.rand(1, 3, 1024, 1024)
model = MAE()
out = model(inp)
print(out[0][0].shape, out[0][1].shape, out[0][2].shape)