Matías Haeussler

@LZDSJTU The FPS implemented in tf_sampling_g.cu is in C++ CUDA because it is used by PointNet++ Set Abstraction layers as a Tensorflow custom operation (https://medium.com/@taxfromdk/writing-a-new-tensorflow-operation-including-c-cuda-forward-gradient-and-grad-check-3c46708351e7), the ones must be implemented...

Hi @Yang7879 , thanks, what you say is probably the issue. 1. How is defined a valid instance? 2. "Invalid blocks" should be skiped from addition to h5 files or...

First I would like to give some context to this issue. It applies only when a different WorkerX/GPUX/etcX tf.name_scope() was created over different instantations of any model that uses tf.train.ExponentialMovingAverage...

Thougth it again and realized that trainable parámeters of Batch Normalization can be defined with tf.get_variable() and use "with tf.variable_scope(tf.get_variable_scope(), reuse=tf.AUTO_REUSE)" as @shimafoolad says, as follows: ````` def batch_norm_template(inputs, is_training,...

> I found a solution which might work also for you > > ``` > while (field.acroField.getWidgets().length) { > field.acroField.removeWidget(0); > } > form.removeField(field); > ``` Thanks This allowed me...

Indeed, the image paths given in "fpath" is joined from behind in light_head_rcnn/experiments/lizming/lighthead[...]/dataset.py this way: `os.path.join(train_root_folder, record['fpath])` where train_root_folder is specified in light_head_rcnn/experiments/lizming/lighthead[...]/config.py: `train_root_folder = os.path.join(root_dir, 'data/MSCOCO')` Finally, root_dir also...

I solved it adjusting "nr_dataflow" in config.py (in the corresponding folder you should be training according to README.md) from 16 to 2 in case of 1 GPU, because train_batch_per_gpu=2, (so...