entalent

error calling `pack_padded_sequence` when using pytorch 1.7

1

Hi, I ran the `tools/train.py` using python 3.7 and torch 1.7, and the program throws the following exception: ``` ...... File "/home/self-critical.pytorch-master/captioning/models/AttModel.py", line 34, in sort_pack_padded_sequence tmp = pack_padded_sequence(input[indices], sorted_lengths,...

Hi, How is the result in README.md acquired? Is oracle evaluation and consensus re-ranking evaluation in section 4.2 of the paper implemented in this repository?

Tips for Ampere GPU users (RTX 3090, 3080/ti, etc.) to run scene graph benchmark with CUDA 11

7

Since Ampere GPUs only supports PyTorch built with CUDA 11 and later, we can only build this repository with CUDA 11.1 and PyTorch 1.8.1 or later. Here are the differences...

tf.split() and tf.concat() parameter order

1

Initializing network with tensorflow 1.2.1 leads to such error: ``` Traceback (most recent call last): File "main.py", line 5, in net = CaffeNet({'data': input_data}) File "/home/caffe-tensorflow/kaffe/tensorflow/network.py", line 47, in __init__...

loss计算的问题

2

在 tf_policy_value_net.py 第54行，定义了每个动作的概率是 self.action_probs = tf.nn.softmax(policy_net_out, name="policy_net_proba")， 第90行定义的loss是 cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=self.action_probs, labels=self.mcts_probs)。 但tensorflow[文档](https://www.tensorflow.org/api_docs/python/tf/nn/softmax_cross_entropy_with_logits)对 tf.nn.softmax_cross_entropy 这个函数有个warning ，是说不要把softmax的输出给这个函数，因为这个函数内部会自己对logits做一遍softmax……所以这里直接用这个loss是不是不太好 另外，请问model文件夹里的模型是从0开始用这份代码训练的，还是已经训练好的theano/pytorch的模型直接转成tensorflow模型得到的？