tejaskhot

# argmax is not differentiable, so the hack to get argmax is
# softmax(x)^T * range(indices)

# numpy
beta = 12
y_est = np.array([[1.1, 3.0, 1.1, 1.3, 0.8]])
# multiplying by some large constant beta to make the resulting
# distribution more peaky near the max
a = np.exp(beta*y_est)
b = np.sum(np.exp(beta*y_est))

tejaskhot

import faiss

# example data
xb = np.random.rand(200000, 32).astype('float32')
xq = np.random.rand(500, 32).astype('float32')

# get reference result with index
index = faiss.IndexFlatL2(xb.shape[1])
index.add(xb)
lims, D, I = index.range_search(xq, 1.5)

tejaskhot

class FocalLoss(nn.Module):
    def __init__(self, alpha=1, gamma=2, logits=False, reduce=True):
        super(FocalLoss, self).__init__()
        self.alpha = alpha
        self.gamma = gamma
        self.logits = logits
        self.reduce = reduce

    def forward(self, inputs, targets):
        if self.logits:

tejaskhot

Project Title

One Paragraph of project description goes here

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment for notes on how to deploy the project on a live system.

Prerequisites

tejaskhot

# This would replaces names of all files in folder having the string ".obj.ply" with ".ply"
for filename in *; do newname=`echo $filename | sed 's/.obj.ply/.ply/g'`; mv $filename $newname; done

tejaskhot

classnames = {
    0:'ceiling',
    1:'floor',
    2:'wall',
    3:'beam',
    4:'column',
    5:'window',
    6:'door',
    7:'table',
    8:'chair',

tejaskhot

Source:
https://stackoverflow.com/questions/2853334/glueing-tile-images-together-using-imagemagicks-montage-command-without-resizin

montage -mode concatenate -tile 12x *.jpg out.jpg

tejaskhot

import torch


def chamfer_distance_without_batch(p1, p2, debug=False):

    '''
    Calculate Chamfer Distance between two point sets
    :param p1: size[1, N, D]
    :param p2: size[1, M, D]
    :param debug: whether need to output debug info

tejaskhot

def exp_lr_scheduler(optimizer, global_step, init_lr, decay_steps, decay_rate, lr_clip, staircase=True):
    """Decay learning rate by a factor of 0.1 every lr_decay_epoch epochs."""
    if staircase:
        lr = init_lr * decay_rate**(global_step // decay_steps)
    else:
        lr = init_lr * decay_rate**(global_step / decay_steps)
    lr = max(lr, lr_clip)

    if global_step % decay_steps == 0:
        print('LR is set to {}'.format(lr))

tejaskhot

# pytorch function to replicate tensorflow's tf.nn.softmax_cross_entropy_with_logits
# works for soft targets or one-hot encodings

import torch
import torch.nn.functional as F

logits = model(input)
loss = torch.sum(- target * F.log_softmax(logits, -1), -1)
mean_loss = loss.mean()