nagadomi

# nunif/iw3 extension for Sapiens2: https://github.com/facebookresearch/sapiens2
# 
# Sapiens2 is a suite of human-centric vision models that includes depth estimation and segmentation,
# allowing the creation of passthrough 3D videos with only people extracted from standard video footage.
#
# Save this file as `iw3/depth_models/_sapiens2_model.py`.
# Checkpoint files (pointmap and seg) can be downloaded from: https://huggingface.co/collections/facebook/sapiens2
# Example:
#   iw3/pretrained_models/hub/checkpoints/sapiens2_0.4b_pointmap.safetensors
#   iw3/pretrained_models/hub/checkpoints/sapiens2_0.4b_seg.safetensors

nagadomi

import ctypes
import os
import sys
import torch
import numpy as np
import torchvision

# NVOF Constants
NV_OF_API_MAJOR_VERSION = 2
NV_OF_API_MINOR_VERSION = 0

nagadomi

import torch
import coremltools as ct
import os

# --- Model Definitions (Primary Models) ---

V2_MODELS = {
    # model_name: (entry_point, arg_name, encoder_name)
    "Any_S": ("DepthAnything", "encoder", "vits"),
    "Any_B": ("DepthAnything", "encoder", "vitb"),

nagadomi

# python generate_person_mask_sam3.py -i ./test_images -o output
#
# This script generates human masks from images.
# It works not only with photos but also with anime-style images.
# The latest version of transformers is required. 
# `pip install git+https://github.com/huggingface/transformers.git`
# Using SAM3 requires proper license authorization.

import os
from os import path

nagadomi

#version 450

layout (location = POSITION) in vec3 aVertexPosition;
layout (location = TEXCOORD0) in vec2 aVertexTextureCoord;
layout (location = TEXCOORD0) out vec2 vTexCoord;

void main() {
    vTexCoord = aVertexTextureCoord;
    gl_Position = vec4(aVertexPosition, 1);
}

nagadomi

#!/bin/bash

export NODE_OPTIONS=""
git submodule sync --recursive
git submodule update --init --recursive

BUILD_DIR=build_jsep_st
rm -fr ${BUILD_DIR}

cd cmake/external/emsdk

nagadomi

import numpy as np
from timeit import timeit 
from collections import deque

# online running meanstd. ref https://github.com/ajcr/rolling/tree/master/rolling
class TimeRunningMeanStd(object):
    def __init__(self, shape, ttl):
        self.ttl = ttl
        self.mean = np.zeros(shape, dtype=np.float64)
        self.std = np.ones(shape, dtype=np.float64)

nagadomi

require 'nn'
local ScaleTable, parent = torch.class("nn.ScaleTable", "nn.Module")

function ScaleTable:__init()
   parent.__init(self)
   self.gradInput = {}
   self.grad_tmp = torch.Tensor()
   self.scale = torch.Tensor()
end
function ScaleTable:updateOutput(input)

nagadomi

# pip3 install python-dateutil websocket-client
import sys
import time
import json
import websocket
from dateutil.parser import parse as date_parse
from collections import deque

K=100
pack_buf = deque(maxlen=K)

nagadomi

# -*- coding: utf-8 -*-

# pip3 install xlrd python-dateutil

import xlrd
from datetime import datetime
from dateutil.parser import parse as date_parse
from collections import defaultdict

# 入力ファイル