guxing119

Results 4 comments of guxing119

博主你好,对于bondingbox的预测,各个有效层卷积后的输出channel数是num_priors*4,这里的4为什么是[x_offset、y_offset、h和w的调整情况], 而不是[预测框的x1, y1, x2, y2(预测框的左上角和右下角的坐标值)]呢,谢谢!

博主你好,感谢你的回答,在代码的ssd_layers.py的最后,先验框产生完之后,为什么还要给prior_boxes增加一个维度,并让它在这个维度上重复(tf.shape(x)[0] => 应该是feature map的宽度吧)次呢 代码: prior_boxes = np.concatenate((prior_boxes, variances), axis=1) prior_boxes_tensor = K.expand_dims(K.variable(prior_boxes), 0) pattern = [tf.shape(x)[0], 1, 1] prior_boxes_tensor = tf.tile(prior_boxes_tensor, pattern) return prior_boxes_tensor 谢谢!

楼主你好,果然是版本问题, 开始我在mac上跑的,查了查,mac最大支持到2.0,换了个环境,就可以了,添麻烦了,多谢! 2020-07-27 11:07:26.007587: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties: pciBusID: 0000:00:04.0 name: Tesla T4 computeCapability: 7.5 coreClock: 1.59GHz coreCount: 40 deviceMemorySize: 14.73GiB deviceMemoryBandwidth: 298.08GiB/s 2020-07-27 11:07:26.007629: I tensorflow/stream_executor/platform/default/dso_loader.cc:44]...