I can't use pb file

[A-Nariman]

I trained my model from scratch without pre-trained weights. Using pb file cannot detect the objects even with threshold 0.1 while the checkpoints are fine and the model can detect the objects perfectly.

I read the all of issues about pb file but there was no solution. I would like to know what the problem is? Is it Darknet or Tensorflow? or something else?

I used the following commands:

1. Training:

---

flow --model cfg/train.cfg --train --annotation train/xml --dataset train/png

2. Saving pb file:

---

flow --model cfg/train.cfg --load -1 --savepb

3. Testing with checkpoints:

---

flow --imgdir sample_img/ --model cfg/train.cfg --load -1

4. Testing with pb file:

---

flow --pbLoad built_graph/train.pb --metaLoad built_graph/train.meta --imgdir sample_img/

[zhangwenwen]

solved？

[A-Nariman]

Yes, I trained my model with a pre-trained weight and worked. Without pre-trained weight does not work!!!

[braunyyy]

I've got the same problem, have you got an idea how to fix this?

[jkstyle2]

Can we not train without any pretrained weights? I've tried to predict some objects after training from the scratch, but I continue to fail detecting. Any suggestion?

[amankumarjain]

Yes, I trained my model with a pre-trained weight and worked. Without pre-trained weight does not work!!!

how to do without pretrained weight?