I am training a Faster-RCNN(VGG-16 architecture) on INRIA Person dataset. I was trained for 180,000 training steps. But when I evaluate the network, it gives varying result with the same image.
Following are the images
I am not sure why does it give different results for the same set of weights.The network is implemented in caffe.
Any insight into the problem is much appreciated.
Following image shows different network losses
Recently, I also prepared my own dataset to training, and got the similar results as yours.
The following is my experiences and share with you:
Check input format include images and your bounding box csvfile or xml (where always put on Annotation file) whether all bounding box (x1, y1, x2, y2) correct?
Then check roidb/imdb loading python script (put on FasterRCNN/lib/datasets/pascal_roi.py, and maybe yours is inria.py),
make sure _load_xxx_annotation() correctly loaded all bounding box by print bounding_box and filename. Importantly, if your script is copied and modified the pascal_roi.py or any prototype script, please check whether it will saved all roi and image info into cache file, if yes you need to delete that cache file when you change any configure files and re-try.
Finally, make sure that all bounding box correctly generating when network is training (e.g. print minibatch variable to show filename and corresponding x1, y1, x2, y2 shown at FasterRCNN/lib/roi_data_layer/layer.py). If roi generator generate correctly, the bounding box will not differ with your manually select bounding box largely.
Some similar issue may cause this problem as well.
Related
for Forge Autodesk.AEC.LevelsExtension, where is the cut plane? looks like it doesn't follow the view range in Revit?
Appended to what Cyrille mentioned.
Revit's AEC model data will be generated automatically during the Forge translation. The level info is dumping from Revit Level elements, e.g., the name, guid, elevations, extensions of the Revit Level which you can see with Revit API. The Levels extension just uses it to rebuild level ranges.
For example, the cutting range of level 1 will be from level1's elevation to (level2's elevation - a height adjustment) with my research. The height adjustment is to avoid cutting on top of floors to brining a better view. So, it's not following Revit's view range of the floor plan view.
If the levels extension doesn't match your need, you may check out my level section tool. This sample demonstrates the concept of how to create cut planes by levels.
In this extension, floors cut planes are defined based on the bounding box of the floor node in the scene instance tree. Floors are defined in an additional AEC data json file which contains additional information.
i'm new to the autodesk viewer and the others api and I could use some help figuring out what tools are the best to do what I want.
I'm using the autodesk viewer to let users generate 2d views of cut planes, in order to do this I simply use the getScreenshot function from the viewer and save it as a Blueprint in my app.
What I would like to do now is that when the user updates his 3d model, to automatically update my 2d views with the new 3d model.
Currently the only solution I came up with is to store the position of the camera when taking the screenshot and then when the 3d model is updated, have another computer in the background go in the viewer and take the screenshots again at the same location.
This does not seems to be a very elegant solution so I would like to know if there's an alternative, like a way to generate 2d views from an api call or maybe use the Design Automation API with the viewer to take the screenshots ?
Another thing i'm struggling with is getting precise measure of the 2ds views i'm generating, my current solution is to calculate the distance between the camera and the cut plane and then use the fov to get an approximate measure, the formula looks like this :
Math.tan((viewer.getCamera().fov / 2) * Math.PI / 180) * distanceBetweenCameraAndPlaneCut * 4;
but it is very dependent on the user facing the plane cut at a 90° angle and i'm thinking there should be something better to do with the measure tool.
Thanks a lot for your time!
You should be able to run the Viewer on a server using puppeteer (without client-side components) to generate screenshots: https://forge.autodesk.com/blog/running-forge-viewer-headless-chrome-puppeteer
Note: you could also use Design Automation API in order to do something similar, but then you are limited to the file formats the given product (e.g. AutoCAD) supports as input
You could also simply save the state of the Viewer and reset it next time you load the same model in order to take a screenshot of the exact same area using getState()/restoreState(): https://adndevblog.typepad.com/cloud_and_mobile/2015/02/managing-viewer-states-from-the-api.html
Why are you trying to measure the distance between cut plane and camera position? Is that in order to restore the Viewer state/camera? If so, then the solution mentioned in 2. should help
As the title says I am looking to use the Design Automation API to upload an archive of blocks, then insert them all into a drawing with a border.
The positioning of the blocks does not matter they just need to be inserted into the drawing ready to be arranged by an engineer.
Any help or advice on end points or limitations would be great thanks.
This is doable. You need to develop a custom activity with 2 inputs and one output. input#1 is the base drawing with boarder and input#2 is a zip file with the blocks. When you submit your workitem you will mark the second input argument as ResourceKind = ResourceKind.ZipPackage this will tell the service that it should unzip the file into the folder designated by LocalFileName. Then your script can enumerate the files in the folder (see vl-directory-files) and issue the INSERT command.
digits 4.0 0.14.0-rc.3 /Ubuntu (aws)
training a 5 class GoogLenet model with about 800 training samples in each class. I was trying to use the bvlc_imagent as pre-trained model. These are the steps I took:
downloaded imagenet from http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel and placed it in /home/ubuntu/models
2.
a. Pasted the "train_val.prototxt" from here https://github.com/BVLC/caffe/blob/master/models/bvlc_reference_caffenet/train_val.prototxt into the custom network tab and
b. '#' commented out the "source" and "backend" lines (since it was complaning about them)
In the pre-trained models text box pasted the path to the '.caffemodel'. in my case: "/home/ubuntu/models/bvlc_googlenet.caffemodel"
I get this error:
ERROR: Cannot copy param 0 weights from layer 'loss1/classifier'; shape mismatch. Source param shape is 1 1 1000 1024 (1024000); target param shape is 6 1024 (6144). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer.
I have pasted various train_val.prototext from github issues etc and no luck unfortunately,
I am not sure why this is getting so complicated, in older versions of digits, we could just enter the path to the folder and it was working great for transfer learning.
Could someone help?
Rename the layer from "loss1/classifier" to "loss1/classifier_retrain".
When fine-tuning a model, here's what Caffe does:
# pseudo-code
for layer in new_model:
if layer.name in old_model:
new_model.layer.weights = old_model.layer.weights
You're getting an error because the weights for "loss1/classifier" were for a 1000-class classification problem (1000x1024), and you're trying to copy them into a layer for a 6-class classification problem (6x1024). When you rename the layer, Caffe doesn't try to copy the weights for that layer and you get randomly initialized weights - which is what you want.
Also, I suggest you use this network description which is already set up as an all-in-one network description for GoogLeNet. It will save you some trouble.
https://github.com/NVIDIA/DIGITS/blob/digits-4.0/digits/standard-networks/caffe/googlenet.prototxt
I'm having trouble filling in my curves using contourf() on Octave. I'm running Octave 3.6.4 on Mac OS X 10.8.5.
When I use contour(x,fp,data3) I get the following, which is correct:
[Contour plot of data, not filled]
However, when I want try contourf(x,fp,data3) in order to fill in the gaps I get this monstrosity:
Contourf plot of data, filled, but not correct
What can I do to fix this? I've read the contourf() documentation and can't see anything there that I'm missing. Any advice would be helpful.
Thanks!
P.S. Here's a link to a smaller version of the data file: https://www.dropbox.com/s/lmvdzi7l42tasr8/Ch942.csv?dl=0. The whole file is huge, so this represents the first few lines but still shows the problem when plotted in Octave.
Unfortunately I don't have enough reputation points to post more than two links, so I've deleted the contour() plot that looks right, but isn't filled in. Sorry.